Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Sorry for being late. I was busy with something else these days. It'll be great to have a dedicated image transferring library that provides both pre-copying and zero-copying sematics, and we are glad to have VMThunder integrated in it. Before that library is done, however, we plan to propose a blue print that solely focus on integrate VMThunder into Open Stack, as a plug-in of course. Then we can move VMThunder into the newly created transferring library with a refactory process. Does this plan make sense? BTW, I'll not be able to goto the summit. It's too far away. Pity. At 2014-04-28 11:01:13,Sheng Bo Hou sb...@cn.ibm.com wrote: Jay, Huiba, Chris, Solly, Zhiyan, and everybody else, I am so excited that two of the proposals: Image Upload Plugin(http://summit.openstack.org/cfp/details/353) and Data transfer service Plugin(http://summit.openstack.org/cfp/details/352) have been merged together and scheduled in the coming design summit. If you show up in Atlanta, please come this session(http://junodesignsummit.sched.org/event/c00119362c07e4cb203d1c4053add187) and start our discussion, on Wednesday, May 14 • 11:50am - 12:30pm. I will propose a common image transfer library for all the OpenStack projects to to upload and download the images. If it is approved, with this library, Huiba, you can feel free to implement the transfer protocols you like. Best wishes, Vincent Hou (侯胜博) Staff Software Engineer, Open Standards and Open Source Team, Emerging Technology Institute, IBM China Software Development Lab Tel: 86-10-82450778 Fax: 86-10-82453660 Notes ID: Sheng Bo Hou/China/IBM@IBMCNE-mail: sb...@cn.ibm.com Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang West Road, Haidian District, Beijing, P.R.C.100193 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层 邮编:100193 | Sheng Bo Hou/China/IBM@IBMCN 2014/04/27 22:33 | Please respond to OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.org | | | To | OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.org, | | cc | | | Subject | Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder | | | | | I have done a little test for the image download and upload. I created an API for the image access, containing copyFrom and sendTo. I moved the image download and upload code from XenApi into the implementation for Http with some modifications, and the code worked for libvirt as well. copyFrom means to download the image and return the image data, and different hypervisors can choose to save it in a file or import it to the datastore; sendTo is used to upload the image and the image data is passed in as a parameter. I also did an investigation about how each hypervisor is doing the image upload and download. For the download: libvirt, hyper-v and baremetal use the code image_service.download to download the image and save it into a file. vmwareapi uses the code image_service.download to download the image and import it into the datastore. XenAPi uses image_service.download to download the image for VHD image. For the upload: They use image_service.upload to upload the image. I think we can conclude that it is possible to have a common image transfer library with different implementations for different protocols. This is a small demo code for the library: https://review.openstack.org/#/c/90601/(Jay, is it close to the library as you mentioned?). I just replaced the upload and download part with the http implementation for the imageapi and it worked fine. Best wishes, Vincent Hou (侯胜博) Staff Software Engineer, Open Standards and Open Source Team, Emerging Technology Institute, IBM China Software Development Lab Tel: 86-10-82450778 Fax: 86-10-82453660 Notes ID: Sheng Bo Hou/China/IBM@IBMCNE-mail: sb...@cn.ibm.com Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang West Road, Haidian District, Beijing, P.R.C.100193 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层 邮编:100193 | Solly Ross sr...@redhat.com 2014/04/25 01:46 | Please respond to OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.org | | | To | OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org, | | cc | | | Subject | Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder | | | | | Something to be aware of when planing an image transfer library is that individual drivers might have optimized support for image transfer in certain cases (especially when dealing with transferring between different formats, like raw to qcow2, etc). This builds on what Christopher was saying -- there's actually a reason why we have code for each driver. While having a common image copying library would be nice, I think a better way to do it would be to have some sort of
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Waiting on review.openstack.org to come back live to look at the demo code and provide more accurate feedback… Interesting and good to hear the code moved easily. The possibility to have a functional a common image transfer service wasn't questioned (IMHO). What I was stating was that we'll need strong data to point to how the common code doesn't degrade download performance for various driver/deployments. I do think having a common set of configuration (and driver calls?) for the download options makes a lot of sense (like glance has done for image_service.download). I'm just a little more cautious when it comes to true common download code at this point. Christopher From: Sheng Bo Hou sb...@cn.ibm.commailto:sb...@cn.ibm.com Reply-To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org Date: Sunday, April 27, 2014 9:33 AM To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder I have done a little test for the image download and upload. I created an API for the image access, containing copyFrom and sendTo. I moved the image download and upload code from XenApi into the implementation for Http with some modifications, and the code worked for libvirt as well. copyFrom means to download the image and return the image data, and different hypervisors can choose to save it in a file or import it to the datastore; sendTo is used to upload the image and the image data is passed in as a parameter. I also did an investigation about how each hypervisor is doing the image upload and download. For the download: libvirt, hyper-v and baremetal use the code image_service.download to download the image and save it into a file. vmwareapi uses the code image_service.download to download the image and import it into the datastore. XenAPi uses image_service.download to download the image for VHD image. For the upload: They use image_service.upload to upload the image. I think we can conclude that it is possible to have a common image transfer library with different implementations for different protocols. This is a small demo code for the library: https://review.openstack.org/#/c/90601/(Jay, is it close to the library as you mentioned?). I just replaced the upload and download part with the http implementation for the imageapi and it worked fine. Best wishes, Vincent Hou (侯胜博) Staff Software Engineer, Open Standards and Open Source Team, Emerging Technology Institute, IBM China Software Development Lab Tel: 86-10-82450778 Fax: 86-10-82453660 Notes ID: Sheng Bo Hou/China/IBM@IBMCNE-mail: sb...@cn.ibm.commailto:sb...@cn.ibm.com Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang West Road, Haidian District, Beijing, P.R.C.100193 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层邮编:100193 Solly Ross sr...@redhat.commailto:sr...@redhat.com 2014/04/25 01:46 Please respond to OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org To OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org, cc Subject Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder Something to be aware of when planing an image transfer library is that individual drivers might have optimized support for image transfer in certain cases (especially when dealing with transferring between different formats, like raw to qcow2, etc). This builds on what Christopher was saying -- there's actually a reason why we have code for each driver. While having a common image copying library would be nice, I think a better way to do it would be to have some sort of library composed of building blocks, such that each driver could make use of common functionality while still tailoring the operation to the quirks of the particular drivers. Best Regards, Solly Ross - Original Message - From: Christopher Lefelhocz christopher.lefel...@rackspace.commailto:christopher.lefel...@rackspace.com To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.orgmailto:openstack-dev@lists.openstack.org Sent: Thursday, April 24, 2014 11:17:41 AM Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder Apologies for coming to this discussion late... On 4/22/14 6:21 PM, Jay Pipes jaypi...@gmail.commailto:jaypi...@gmail.com wrote: Right, a good solution would allow for some flexibility via multiple transfer drivers. +1. In particular I don't think this discussion should degenerate into zero-copy vs. pre caching. I see both as
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
I have done a little test for the image download and upload. I created an API for the image access, containing copyFrom and sendTo. I moved the image download and upload code from XenApi into the implementation for Http with some modifications, and the code worked for libvirt as well. copyFrom means to download the image and return the image data, and different hypervisors can choose to save it in a file or import it to the datastore; sendTo is used to upload the image and the image data is passed in as a parameter. I also did an investigation about how each hypervisor is doing the image upload and download. For the download: libvirt, hyper-v and baremetal use the code image_service.download to download the image and save it into a file. vmwareapi uses the code image_service.download to download the image and import it into the datastore. XenAPi uses image_service.download to download the image for VHD image. For the upload: They use image_service.upload to upload the image. I think we can conclude that it is possible to have a common image transfer library with different implementations for different protocols. This is a small demo code for the library: https://review.openstack.org/#/c/90601/(Jay, is it close to the library as you mentioned?). I just replaced the upload and download part with the http implementation for the imageapi and it worked fine. Best wishes, Vincent Hou (侯胜博) Staff Software Engineer, Open Standards and Open Source Team, Emerging Technology Institute, IBM China Software Development Lab Tel: 86-10-82450778 Fax: 86-10-82453660 Notes ID: Sheng Bo Hou/China/IBM@IBMCNE-mail: sb...@cn.ibm.com Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang West Road, Haidian District, Beijing, P.R.C.100193 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层 邮编:100193 Solly Ross sr...@redhat.com 2014/04/25 01:46 Please respond to OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.org To OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org, cc Subject Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder Something to be aware of when planing an image transfer library is that individual drivers might have optimized support for image transfer in certain cases (especially when dealing with transferring between different formats, like raw to qcow2, etc). This builds on what Christopher was saying -- there's actually a reason why we have code for each driver. While having a common image copying library would be nice, I think a better way to do it would be to have some sort of library composed of building blocks, such that each driver could make use of common functionality while still tailoring the operation to the quirks of the particular drivers. Best Regards, Solly Ross - Original Message - From: Christopher Lefelhocz christopher.lefel...@rackspace.com To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org Sent: Thursday, April 24, 2014 11:17:41 AM Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder Apologies for coming to this discussion late... On 4/22/14 6:21 PM, Jay Pipes jaypi...@gmail.com wrote: Right, a good solution would allow for some flexibility via multiple transfer drivers. +1. In particular I don't think this discussion should degenerate into zero-copy vs. pre caching. I see both as possible solutions depending on deployer/environment needs. Jay Pipes has suggested we figure out a blueprint for a separate library dedicated to the data(byte) transfer, which may be put in oslo and used by any projects in need (Hoping Jay can come in:-)). Huiba, Zhiyan, everyone else, do you think we come up with a blueprint about the data transfer in oslo can work? Yes, so I believe the most appropriate solution is to create a library -- in oslo or a standalone library like taskflow -- that would offer a simple byte streaming library that could be used by nova.image to expose a neat and clean task-based API. Right now, there is a bunch of random image transfer code spread throughout nova.image and in each of the virt drivers there seems to be different re-implementations of similar functionality. I propose we clean all that up and have nova.image expose an API so that a virt driver could do something like this: from nova.image import api as image_api ... task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() Within nova.image.api.copy(), we would use the aforementioned transfer library to move the image bits from the source to the destination using the most appropriate method. If I understand correctly, we'll create some common library around this. It would be good to understand the details a bit better. I've thought a bit about this issue.
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Jay, Huiba, Chris, Solly, Zhiyan, and everybody else, I am so excited that two of the proposals: Image Upload Plugin( http://summit.openstack.org/cfp/details/353) and Data transfer service Plugin(http://summit.openstack.org/cfp/details/352) have been merged together and scheduled in the coming design summit. If you show up in Atlanta, please come this session( http://junodesignsummit.sched.org/event/c00119362c07e4cb203d1c4053add187) and start our discussion, on Wednesday, May 14 • 11:50am - 12:30pm. I will propose a common image transfer library for all the OpenStack projects to to upload and download the images. If it is approved, with this library, Huiba, you can feel free to implement the transfer protocols you like. Best wishes, Vincent Hou (侯胜博) Staff Software Engineer, Open Standards and Open Source Team, Emerging Technology Institute, IBM China Software Development Lab Tel: 86-10-82450778 Fax: 86-10-82453660 Notes ID: Sheng Bo Hou/China/IBM@IBMCNE-mail: sb...@cn.ibm.com Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang West Road, Haidian District, Beijing, P.R.C.100193 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层 邮编:100193 Sheng Bo Hou/China/IBM@IBMCN 2014/04/27 22:33 Please respond to OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.org To OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.org, cc Subject Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder I have done a little test for the image download and upload. I created an API for the image access, containing copyFrom and sendTo. I moved the image download and upload code from XenApi into the implementation for Http with some modifications, and the code worked for libvirt as well. copyFrom means to download the image and return the image data, and different hypervisors can choose to save it in a file or import it to the datastore; sendTo is used to upload the image and the image data is passed in as a parameter. I also did an investigation about how each hypervisor is doing the image upload and download. For the download: libvirt, hyper-v and baremetal use the code image_service.download to download the image and save it into a file. vmwareapi uses the code image_service.download to download the image and import it into the datastore. XenAPi uses image_service.download to download the image for VHD image. For the upload: They use image_service.upload to upload the image. I think we can conclude that it is possible to have a common image transfer library with different implementations for different protocols. This is a small demo code for the library: https://review.openstack.org/#/c/90601/(Jay, is it close to the library as you mentioned?). I just replaced the upload and download part with the http implementation for the imageapi and it worked fine. Best wishes, Vincent Hou (侯胜博) Staff Software Engineer, Open Standards and Open Source Team, Emerging Technology Institute, IBM China Software Development Lab Tel: 86-10-82450778 Fax: 86-10-82453660 Notes ID: Sheng Bo Hou/China/IBM@IBMCNE-mail: sb...@cn.ibm.com Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang West Road, Haidian District, Beijing, P.R.C.100193 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层 邮编:100193 Solly Ross sr...@redhat.com 2014/04/25 01:46 Please respond to OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.org To OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org, cc Subject Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder Something to be aware of when planing an image transfer library is that individual drivers might have optimized support for image transfer in certain cases (especially when dealing with transferring between different formats, like raw to qcow2, etc). This builds on what Christopher was saying -- there's actually a reason why we have code for each driver. While having a common image copying library would be nice, I think a better way to do it would be to have some sort of library composed of building blocks, such that each driver could make use of common functionality while still tailoring the operation to the quirks of the particular drivers. Best Regards, Solly Ross - Original Message - From: Christopher Lefelhocz christopher.lefel...@rackspace.com To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org Sent: Thursday, April 24, 2014 11:17:41 AM Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder Apologies for coming to this discussion late... On 4/22/14 6:21 PM, Jay Pipes jaypi...@gmail.com wrote: Right, a good solution would allow for some flexibility
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Hmm, I totally see the value of doing this. Not sure that there could be the same kinds of liveness guarantees with non-shared-storage, but I am certainly happy to see a proof of concept in this area! :) By liveness, if you mean down time of migration, our current results show that liveness is guaranteed with non-shared-storage. Some preliminary work has been published in a conference SOSE14, which can be found at http://www.vmthunder.org/dlsm_sose2014_final.pdf And we have made some improvements to it, and the work is still under development. We are planning to write a new paper and submit it to another conference in this summer. how about zero-copying? It would be an implementation detail within nova.image.api.copy() function (and the aforementioned image bits mover library) :) IMHO, (pre-)copying and zero-copying are different in nature, and it's not necessary to mask such difference by a single interface. With 2 sets of interfaces, programmers (users of copying service) will be reminded of the cost of (pre-)copying, or the risk of runtime network congestion of zero-copying. At 2014-04-23 23:02:29,Jay Pipes jaypi...@gmail.com wrote: On Wed, 2014-04-23 at 13:56 +0800, lihuiba wrote: For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. With a good zero-copy transfer lib, live migration support can be extended to non-shared storage, or cross-datacenter. It's a kind of value. Hmm, I totally see the value of doing this. Not sure that there could be the same kinds of liveness guarantees with non-shared-storage, but I am certainly happy to see a proof of concept in this area! :) task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() +1 looks cool! how about zero-copying? It would be an implementation detail within nova.image.api.copy() function (and the aforementioned image bits mover library) :) The key here is to leak as little implementation detail out of the nova.image.api module Best, -jay At 2014-04-23 07:21:27,Jay Pipes jaypi...@gmail.com wrote: Hi Vincent, Zhi, Huiba, sorry for delayed response. See comments inline. On Tue, 2014-04-22 at 10:59 +0800, Sheng Bo Hou wrote: I actually support the idea Huiba has proposed, and I am thinking of how to optimize the large data transfer(for example, 100G in a short time) as well. I registered two blueprints in nova-specs, one is for an image upload plug-in to upload the image to glance(https://review.openstack.org/#/c/84671/), the other is a data transfer plug-in(https://review.openstack.org/#/c/87207/) for data migration among nova nodes. I would like to see other transfer protocols, like FTP, bitTorrent, p2p, etc, implemented for data transfer in OpenStack besides HTTP. Data transfer may have many use cases. I summarize them into two catalogs. Please feel free to comment on it. 1. The machines are located in one network, e.g. one domain, one cluster, etc. The characteristic is the machines can access each other directly via the IP addresses(VPN is beyond consideration). In this case, data can be transferred via iSCSI, NFS, and definitive zero-copy as Zhiyan mentioned. 2. The machines are located in different networks, e.g. two data centers, two firewalls, etc. The characteristic is the machines can not access each other directly via the IP addresses(VPN is beyond consideration). The machines are isolated, so they can not be connected with iSCSI, NFS, etc. In this case, data have to go via the protocols, like HTTP, FTP, p2p, etc. I am not sure whether zero-copy can work for this case. Zhiyan, please help me with this doubt. I guess for data transfer, including image downloading, image uploading, live migration, etc, OpenStack needs to taken into account the above two catalogs for data transfer. For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. It is hard to say that one protocol is better than another, and one approach prevails another(BitTorrent is very cool, but if there is only one source and only one target, it would not be that faster than a direct FTP). The key is the use case(FYI:http://amigotechnotes.wordpress.com/2013/12/23/file-transmission-with-different-sharing-solution-on-nas/). Right, a good solution would allow for some flexibility via multiple transfer drivers. Jay Pipes has suggested we figure out a blueprint for a separate library dedicated to the data(byte) transfer, which may be put in oslo and used by any projects in need (Hoping Jay can come in:-)). Huiba, Zhiyan, everyone else, do you think we come up with a blueprint about the data transfer in oslo can work? Yes, so I believe the most appropriate solution is to create a library -- in oslo or a standalone library like taskflow --
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
First, after the discussing between I and Vincent, I'm sure what we talked for zero-copying in above mail threads is different with full-copying/transferring. The transferring/full-copying means image bits duplication, it focuses on using some method to accelerate image bits replication and transferring between Glance/backend-storage and Nova compute host, like P2P, FTP, HTTP, etc., but zero-copying means there is NO any bits duplication and transferring happened between those two sides during the image preparing for VM provisioning, so the latter one focuses on a) how to attaching remote image volume/disk within Glance managed backend-storage from Nova compute host directly, and b) making the VM's root disk from remote template image based on such hypervisor and particular storage technology, btw c) preventing image bits uploading for VM snapshot/capture case. They are totally different to me. (refer: review comments in https://review.openstack.org/#/c/84671/ ) Second, on the implementation level, I consider to put all image handling related code into nova.image namespace sounds neat but I think it can not work (the leak as last mail side here). IMO, the transferring/full-copying logic is more applicable for nova.image namespace, such transferring approach can be implemented based on existing download module plugins structure, e.g. P2P, FTP, but for the zero-copying, regarding to my above points of view, I consider to implement it in nova.virt + nova.virt.hypervisor is make more sense, since it's more related with particular hypervisor and/or storage technology. (refer: inline comments in https://review.openstack.org/#/c/86583/6/specs/juno/image-multiple-location.rst ) zhiyan On Wed, Apr 23, 2014 at 11:02 PM, Jay Pipes jaypi...@gmail.com wrote: On Wed, 2014-04-23 at 13:56 +0800, lihuiba wrote: For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. With a good zero-copy transfer lib, live migration support can be extended to non-shared storage, or cross-datacenter. It's a kind of value. Hmm, I totally see the value of doing this. Not sure that there could be the same kinds of liveness guarantees with non-shared-storage, but I am certainly happy to see a proof of concept in this area! :) task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() +1 looks cool! how about zero-copying? It would be an implementation detail within nova.image.api.copy() function (and the aforementioned image bits mover library) :) The key here is to leak as little implementation detail out of the nova.image.api module Best, -jay At 2014-04-23 07:21:27,Jay Pipes jaypi...@gmail.com wrote: Hi Vincent, Zhi, Huiba, sorry for delayed response. See comments inline. On Tue, 2014-04-22 at 10:59 +0800, Sheng Bo Hou wrote: I actually support the idea Huiba has proposed, and I am thinking of how to optimize the large data transfer(for example, 100G in a short time) as well. I registered two blueprints in nova-specs, one is for an image upload plug-in to upload the image to glance(https://review.openstack.org/#/c/84671/), the other is a data transfer plug-in(https://review.openstack.org/#/c/87207/) for data migration among nova nodes. I would like to see other transfer protocols, like FTP, bitTorrent, p2p, etc, implemented for data transfer in OpenStack besides HTTP. Data transfer may have many use cases. I summarize them into two catalogs. Please feel free to comment on it. 1. The machines are located in one network, e.g. one domain, one cluster, etc. The characteristic is the machines can access each other directly via the IP addresses(VPN is beyond consideration). In this case, data can be transferred via iSCSI, NFS, and definitive zero-copy as Zhiyan mentioned. 2. The machines are located in different networks, e.g. two data centers, two firewalls, etc. The characteristic is the machines can not access each other directly via the IP addresses(VPN is beyond consideration). The machines are isolated, so they can not be connected with iSCSI, NFS, etc. In this case, data have to go via the protocols, like HTTP, FTP, p2p, etc. I am not sure whether zero-copy can work for this case. Zhiyan, please help me with this doubt. I guess for data transfer, including image downloading, image uploading, live migration, etc, OpenStack needs to taken into account the above two catalogs for data transfer. For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. It is hard to say that one protocol is better than another, and one approach prevails another(BitTorrent is very cool, but if there is only one source and only one target, it would not be that faster than a direct FTP). The key is the use
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Right, our discussion ends up with two directions to access the image: full-copying(upload and download the whole image) and zero-copying(remotely access the image and copy on demand). For image transfer via full-copying, nova.image module works, because we only care about how the data(bytes/image) is going to send. It does not matter what the hypervisor is and what the storage it is. I think we can take the library approach with task = image_api.copy(from_path_or_uri, to_path_or_uri) as Jay proposed. For image transfer via zero-copying as Zhi Yan mentioned, depends on the hypervisor and the backend storage. nova.virt is able to access hypervisor and storage context, but not nova.image. The image does not have to be in the same location, where to launch the VM. We care about how the data(image) is accessed(how the VM is launched.) Each of them needs one module to put, and it is hard to merge them. Best wishes, Vincent Hou (侯胜博) Staff Software Engineer, Open Standards and Open Source Team, Emerging Technology Institute, IBM China Software Development Lab Tel: 86-10-82450778 Fax: 86-10-82453660 Notes ID: Sheng Bo Hou/China/IBM@IBMCNE-mail: sb...@cn.ibm.com Address:3F Ring, Building 28 Zhongguancun Software Park, 8 Dongbeiwang West Road, Haidian District, Beijing, P.R.C.100193 地址:北京市海淀区东北旺西路8号中关村软件园28号楼环宇大厦3层 邮编:100193 Zhi Yan Liu lzy@gmail.com 2014/04/24 14:30 Please respond to OpenStack Development Mailing List \(not for usage questions\) openstack-dev@lists.openstack.org To OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org, cc Subject Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder First, after the discussing between I and Vincent, I'm sure what we talked for zero-copying in above mail threads is different with full-copying/transferring. The transferring/full-copying means image bits duplication, it focuses on using some method to accelerate image bits replication and transferring between Glance/backend-storage and Nova compute host, like P2P, FTP, HTTP, etc., but zero-copying means there is NO any bits duplication and transferring happened between those two sides during the image preparing for VM provisioning, so the latter one focuses on a) how to attaching remote image volume/disk within Glance managed backend-storage from Nova compute host directly, and b) making the VM's root disk from remote template image based on such hypervisor and particular storage technology, btw c) preventing image bits uploading for VM snapshot/capture case. They are totally different to me. (refer: review comments in https://review.openstack.org/#/c/84671/ ) Second, on the implementation level, I consider to put all image handling related code into nova.image namespace sounds neat but I think it can not work (the leak as last mail side here). IMO, the transferring/full-copying logic is more applicable for nova.image namespace, such transferring approach can be implemented based on existing download module plugins structure, e.g. P2P, FTP, but for the zero-copying, regarding to my above points of view, I consider to implement it in nova.virt + nova.virt.hypervisor is make more sense, since it's more related with particular hypervisor and/or storage technology. (refer: inline comments in https://review.openstack.org/#/c/86583/6/specs/juno/image-multiple-location.rst ) zhiyan On Wed, Apr 23, 2014 at 11:02 PM, Jay Pipes jaypi...@gmail.com wrote: On Wed, 2014-04-23 at 13:56 +0800, lihuiba wrote: For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. With a good zero-copy transfer lib, live migration support can be extended to non-shared storage, or cross-datacenter. It's a kind of value. Hmm, I totally see the value of doing this. Not sure that there could be the same kinds of liveness guarantees with non-shared-storage, but I am certainly happy to see a proof of concept in this area! :) task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() +1 looks cool! how about zero-copying? It would be an implementation detail within nova.image.api.copy() function (and the aforementioned image bits mover library) :) The key here is to leak as little implementation detail out of the nova.image.api module Best, -jay At 2014-04-23 07:21:27,Jay Pipes jaypi...@gmail.com wrote: Hi Vincent, Zhi, Huiba, sorry for delayed response. See comments inline. On Tue, 2014-04-22 at 10:59 +0800, Sheng Bo Hou wrote: I actually support the idea Huiba has proposed, and I am thinking of how to optimize the large data transfer(for example, 100G in a short time) as well. I registered two blueprints in nova-specs, one is for an image upload plug-in to upload the image to glance(https://review.openstack.org/#/c/84671/), the other is a data
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Apologies for coming to this discussion late... On 4/22/14 6:21 PM, Jay Pipes jaypi...@gmail.com wrote: Right, a good solution would allow for some flexibility via multiple transfer drivers. +1. In particular I don't think this discussion should degenerate into zero-copy vs. pre caching. I see both as possible solutions depending on deployer/environment needs. Jay Pipes has suggested we figure out a blueprint for a separate library dedicated to the data(byte) transfer, which may be put in oslo and used by any projects in need (Hoping Jay can come in:-)). Huiba, Zhiyan, everyone else, do you think we come up with a blueprint about the data transfer in oslo can work? Yes, so I believe the most appropriate solution is to create a library -- in oslo or a standalone library like taskflow -- that would offer a simple byte streaming library that could be used by nova.image to expose a neat and clean task-based API. Right now, there is a bunch of random image transfer code spread throughout nova.image and in each of the virt drivers there seems to be different re-implementations of similar functionality. I propose we clean all that up and have nova.image expose an API so that a virt driver could do something like this: from nova.image import api as image_api ... task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() Within nova.image.api.copy(), we would use the aforementioned transfer library to move the image bits from the source to the destination using the most appropriate method. If I understand correctly, we'll create some common library around this. It would be good to understand the details a bit better. I've thought a bit about this issue. The one area that I get stuck at is providing a common set of downloads which work across drivers effectively. Part of the reason there are a bunch or random image transfers is historical, but also because performance was already a problem. Examples include: transferring to compute first then copying to dom0 causing performance issues, needs in some drivers to download image completely to validate prior to putting in place, etc. It may be easy to say we'll push most of this to the dom0, but I know for Xen our python stack is somewhat limited so that may be an issue. By the way we've been working on proposing a simpler image pre caching system/strategy. It focuses specifically on the image caching portion of this discussion. For those interested, see the nova-spec https://review.openstack.org/#/c/85792. We'd like to leverage whatever optimized image download strategy is available. Christopher ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Something to be aware of when planing an image transfer library is that individual drivers might have optimized support for image transfer in certain cases (especially when dealing with transferring between different formats, like raw to qcow2, etc). This builds on what Christopher was saying -- there's actually a reason why we have code for each driver. While having a common image copying library would be nice, I think a better way to do it would be to have some sort of library composed of building blocks, such that each driver could make use of common functionality while still tailoring the operation to the quirks of the particular drivers. Best Regards, Solly Ross - Original Message - From: Christopher Lefelhocz christopher.lefel...@rackspace.com To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org Sent: Thursday, April 24, 2014 11:17:41 AM Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder Apologies for coming to this discussion late... On 4/22/14 6:21 PM, Jay Pipes jaypi...@gmail.com wrote: Right, a good solution would allow for some flexibility via multiple transfer drivers. +1. In particular I don't think this discussion should degenerate into zero-copy vs. pre caching. I see both as possible solutions depending on deployer/environment needs. Jay Pipes has suggested we figure out a blueprint for a separate library dedicated to the data(byte) transfer, which may be put in oslo and used by any projects in need (Hoping Jay can come in:-)). Huiba, Zhiyan, everyone else, do you think we come up with a blueprint about the data transfer in oslo can work? Yes, so I believe the most appropriate solution is to create a library -- in oslo or a standalone library like taskflow -- that would offer a simple byte streaming library that could be used by nova.image to expose a neat and clean task-based API. Right now, there is a bunch of random image transfer code spread throughout nova.image and in each of the virt drivers there seems to be different re-implementations of similar functionality. I propose we clean all that up and have nova.image expose an API so that a virt driver could do something like this: from nova.image import api as image_api ... task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() Within nova.image.api.copy(), we would use the aforementioned transfer library to move the image bits from the source to the destination using the most appropriate method. If I understand correctly, we'll create some common library around this. It would be good to understand the details a bit better. I've thought a bit about this issue. The one area that I get stuck at is providing a common set of downloads which work across drivers effectively. Part of the reason there are a bunch or random image transfers is historical, but also because performance was already a problem. Examples include: transferring to compute first then copying to dom0 causing performance issues, needs in some drivers to download image completely to validate prior to putting in place, etc. It may be easy to say we'll push most of this to the dom0, but I know for Xen our python stack is somewhat limited so that may be an issue. By the way we've been working on proposing a simpler image pre caching system/strategy. It focuses specifically on the image caching portion of this discussion. For those interested, see the nova-spec https://review.openstack.org/#/c/85792. We'd like to leverage whatever optimized image download strategy is available. Christopher ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations.With a good zero-copy transfer lib, live migration support can be extended to non-shared storage, or cross-datacenter. It's a kind ofvalue. task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() +1 looks cool! how about zero-copying? At 2014-04-23 07:21:27,Jay Pipes jaypi...@gmail.com wrote: Hi Vincent, Zhi, Huiba, sorry for delayed response. See comments inline. On Tue, 2014-04-22 at 10:59 +0800, Sheng Bo Hou wrote: I actually support the idea Huiba has proposed, and I am thinking of how to optimize the large data transfer(for example, 100G in a short time) as well. I registered two blueprints in nova-specs, one is for an image upload plug-in to upload the image to glance(https://review.openstack.org/#/c/84671/), the other is a data transfer plug-in(https://review.openstack.org/#/c/87207/) for data migration among nova nodes. I would like to see other transfer protocols, like FTP, bitTorrent, p2p, etc, implemented for data transfer in OpenStack besides HTTP. Data transfer may have many use cases. I summarize them into two catalogs. Please feel free to comment on it. 1. The machines are located in one network, e.g. one domain, one cluster, etc. The characteristic is the machines can access each other directly via the IP addresses(VPN is beyond consideration). In this case, data can be transferred via iSCSI, NFS, and definitive zero-copy as Zhiyan mentioned. 2. The machines are located in different networks, e.g. two data centers, two firewalls, etc. The characteristic is the machines can not access each other directly via the IP addresses(VPN is beyond consideration). The machines are isolated, so they can not be connected with iSCSI, NFS, etc. In this case, data have to go via the protocols, like HTTP, FTP, p2p, etc. I am not sure whether zero-copy can work for this case. Zhiyan, please help me with this doubt. I guess for data transfer, including image downloading, image uploading, live migration, etc, OpenStack needs to taken into account the above two catalogs for data transfer. For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. It is hard to say that one protocol is better than another, and one approach prevails another(BitTorrent is very cool, but if there is only one source and only one target, it would not be that faster than a direct FTP). The key is the use case(FYI:http://amigotechnotes.wordpress.com/2013/12/23/file-transmission-with-different-sharing-solution-on-nas/). Right, a good solution would allow for some flexibility via multiple transfer drivers. Jay Pipes has suggested we figure out a blueprint for a separate library dedicated to the data(byte) transfer, which may be put in oslo and used by any projects in need (Hoping Jay can come in:-)). Huiba, Zhiyan, everyone else, do you think we come up with a blueprint about the data transfer in oslo can work? Yes, so I believe the most appropriate solution is to create a library -- in oslo or a standalone library like taskflow -- that would offer a simple byte streaming library that could be used by nova.image to expose a neat and clean task-based API. Right now, there is a bunch of random image transfer code spread throughout nova.image and in each of the virt drivers there seems to be different re-implementations of similar functionality. I propose we clean all that up and have nova.image expose an API so that a virt driver could do something like this: from nova.image import api as image_api ... task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() Within nova.image.api.copy(), we would use the aforementioned transfer library to move the image bits from the source to the destination using the most appropriate method. Best, -jay ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
On Wed, 2014-04-23 at 13:56 +0800, lihuiba wrote: For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. With a good zero-copy transfer lib, live migration support can be extended to non-shared storage, or cross-datacenter. It's a kind of value. Hmm, I totally see the value of doing this. Not sure that there could be the same kinds of liveness guarantees with non-shared-storage, but I am certainly happy to see a proof of concept in this area! :) task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() +1 looks cool! how about zero-copying? It would be an implementation detail within nova.image.api.copy() function (and the aforementioned image bits mover library) :) The key here is to leak as little implementation detail out of the nova.image.api module Best, -jay At 2014-04-23 07:21:27,Jay Pipes jaypi...@gmail.com wrote: Hi Vincent, Zhi, Huiba, sorry for delayed response. See comments inline. On Tue, 2014-04-22 at 10:59 +0800, Sheng Bo Hou wrote: I actually support the idea Huiba has proposed, and I am thinking of how to optimize the large data transfer(for example, 100G in a short time) as well. I registered two blueprints in nova-specs, one is for an image upload plug-in to upload the image to glance(https://review.openstack.org/#/c/84671/), the other is a data transfer plug-in(https://review.openstack.org/#/c/87207/) for data migration among nova nodes. I would like to see other transfer protocols, like FTP, bitTorrent, p2p, etc, implemented for data transfer in OpenStack besides HTTP. Data transfer may have many use cases. I summarize them into two catalogs. Please feel free to comment on it. 1. The machines are located in one network, e.g. one domain, one cluster, etc. The characteristic is the machines can access each other directly via the IP addresses(VPN is beyond consideration). In this case, data can be transferred via iSCSI, NFS, and definitive zero-copy as Zhiyan mentioned. 2. The machines are located in different networks, e.g. two data centers, two firewalls, etc. The characteristic is the machines can not access each other directly via the IP addresses(VPN is beyond consideration). The machines are isolated, so they can not be connected with iSCSI, NFS, etc. In this case, data have to go via the protocols, like HTTP, FTP, p2p, etc. I am not sure whether zero-copy can work for this case. Zhiyan, please help me with this doubt. I guess for data transfer, including image downloading, image uploading, live migration, etc, OpenStack needs to taken into account the above two catalogs for data transfer. For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. It is hard to say that one protocol is better than another, and one approach prevails another(BitTorrent is very cool, but if there is only one source and only one target, it would not be that faster than a direct FTP). The key is the use case(FYI:http://amigotechnotes.wordpress.com/2013/12/23/file-transmission-with-different-sharing-solution-on-nas/). Right, a good solution would allow for some flexibility via multiple transfer drivers. Jay Pipes has suggested we figure out a blueprint for a separate library dedicated to the data(byte) transfer, which may be put in oslo and used by any projects in need (Hoping Jay can come in:-)). Huiba, Zhiyan, everyone else, do you think we come up with a blueprint about the data transfer in oslo can work? Yes, so I believe the most appropriate solution is to create a library -- in oslo or a standalone library like taskflow -- that would offer a simple byte streaming library that could be used by nova.image to expose a neat and clean task-based API. Right now, there is a bunch of random image transfer code spread throughout nova.image and in each of the virt drivers there seems to be different re-implementations of similar functionality. I propose we clean all that up and have nova.image expose an API so that a virt driver could do something like this: from nova.image import api as image_api ... task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() Within nova.image.api.copy(), we would use the aforementioned transfer library to move the image bits from the source to the destination using the most appropriate method. Best, -jay ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. Hi, I'm the first author of the TPDS paper of VMThunder. Based on the same techniques and architecture of VMThunder, we have given a new live migration method. As the same as VMThunder, this new live migration method is a storage level solution and can support several kinds of popular hypervisor without modifying them. Technically, this method is a pre-copy and post-copy hybrid method and based on a previous work of ours, which is a live storage migration. (DLSM: Decoupled Live Storage Migration with Distributed Device Mapper Storage. *accepted by t**he 8th IEEE International Symposium on Service-Oriented System Engineering (SOSE)*, Oxford, U.K. April, 2014.) We are now working on this method and preparing to post a new paper to one coming conference. In summary, I think that the architecture of VMThunder is a powerful platform that can support several popular IaaS features, like large scale provisioning, bulk data dissemination, live migration, live spawning, and etc. (VM cluster live spawning is also a current work of us that still under researching) Regards, Thank you! ~-~-~-~-~-~-~-~-~-~-~ Brian Zhaoning Zhang PhD Cand. PDL, NUDT ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Hi Vincent, Zhi, Huiba, sorry for delayed response. See comments inline. On Tue, 2014-04-22 at 10:59 +0800, Sheng Bo Hou wrote: I actually support the idea Huiba has proposed, and I am thinking of how to optimize the large data transfer(for example, 100G in a short time) as well. I registered two blueprints in nova-specs, one is for an image upload plug-in to upload the image to glance(https://review.openstack.org/#/c/84671/), the other is a data transfer plug-in(https://review.openstack.org/#/c/87207/) for data migration among nova nodes. I would like to see other transfer protocols, like FTP, bitTorrent, p2p, etc, implemented for data transfer in OpenStack besides HTTP. Data transfer may have many use cases. I summarize them into two catalogs. Please feel free to comment on it. 1. The machines are located in one network, e.g. one domain, one cluster, etc. The characteristic is the machines can access each other directly via the IP addresses(VPN is beyond consideration). In this case, data can be transferred via iSCSI, NFS, and definitive zero-copy as Zhiyan mentioned. 2. The machines are located in different networks, e.g. two data centers, two firewalls, etc. The characteristic is the machines can not access each other directly via the IP addresses(VPN is beyond consideration). The machines are isolated, so they can not be connected with iSCSI, NFS, etc. In this case, data have to go via the protocols, like HTTP, FTP, p2p, etc. I am not sure whether zero-copy can work for this case. Zhiyan, please help me with this doubt. I guess for data transfer, including image downloading, image uploading, live migration, etc, OpenStack needs to taken into account the above two catalogs for data transfer. For live migration, we use shared storage so I don't think it's quite the same as getting/putting image bits from/to arbitrary locations. It is hard to say that one protocol is better than another, and one approach prevails another(BitTorrent is very cool, but if there is only one source and only one target, it would not be that faster than a direct FTP). The key is the use case(FYI:http://amigotechnotes.wordpress.com/2013/12/23/file-transmission-with-different-sharing-solution-on-nas/). Right, a good solution would allow for some flexibility via multiple transfer drivers. Jay Pipes has suggested we figure out a blueprint for a separate library dedicated to the data(byte) transfer, which may be put in oslo and used by any projects in need (Hoping Jay can come in:-)). Huiba, Zhiyan, everyone else, do you think we come up with a blueprint about the data transfer in oslo can work? Yes, so I believe the most appropriate solution is to create a library -- in oslo or a standalone library like taskflow -- that would offer a simple byte streaming library that could be used by nova.image to expose a neat and clean task-based API. Right now, there is a bunch of random image transfer code spread throughout nova.image and in each of the virt drivers there seems to be different re-implementations of similar functionality. I propose we clean all that up and have nova.image expose an API so that a virt driver could do something like this: from nova.image import api as image_api ... task = image_api.copy(from_path_or_uri, to_path_or_uri) # do some other work copy_task_result = task.wait() Within nova.image.api.copy(), we would use the aforementioned transfer library to move the image bits from the source to the destination using the most appropriate method. Best, -jay ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
On Fri, Apr 18, 2014 at 10:52 PM, lihuiba magazine.lihu...@163.com wrote: btw, I see but at the moment we had fixed it by network interface device driver instead of workaround - to limit network traffic slow down. Which kind of driver, in host kernel, in guest kernel or in openstack? In compute host kernel, doesn't related with OpenStack. There are few works done in Glance (https://blueprints.launchpad.net/glance/+spec/glance-cinder-driver ), but some work still need to be taken I'm sure. There are something on drafting, and some dependencies need to be resolved as well. I read the blueprints carefully, but still have some doubts. Will it store an image as a single volume in cinder? Or store all image Yes files in one shared volume (with a file system on the volume, of course)? Openstack already has support to convert an image to a volume, and to boot from a volume. Are these features similar to this blueprint? Not similar but it could be leverage for this case. I prefer to talk this details in IRC. (And I had read all VMThunder code at today early (my timezone), there are some questions from me as well) zhiyan Huiba Li National Key Laboratory for Parallel and Distributed Processing, College of Computer Science, National University of Defense Technology, Changsha, Hunan Province, P.R. China 410073 At 2014-04-18 12:14:25,Zhi Yan Liu lzy@gmail.com wrote: On Fri, Apr 18, 2014 at 10:53 AM, lihuiba magazine.lihu...@163.com wrote: It's not 100% true, in my case at last. We fixed this problem by network interface driver, it causes kernel panic and readonly issues under heavy networking workload actually. Network traffic control could help. The point is to ensure no instance is starved to death. Traffic control can be done with tc. btw, I see but at the moment we had fixed it by network interface device driver instead of workaround - to limit network traffic slow down. btw, we are doing some works to make Glance to integrate Cinder as a unified block storage backend. That sounds interesting. Is there some more materials? There are few works done in Glance (https://blueprints.launchpad.net/glance/+spec/glance-cinder-driver ), but some work still need to be taken I'm sure. There are something on drafting, and some dependencies need to be resolved as well. At 2014-04-18 06:05:23,Zhi Yan Liu lzy@gmail.com wrote: Replied as inline comments. On Thu, Apr 17, 2014 at 9:33 PM, lihuiba magazine.lihu...@163.com wrote: IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Yes, in this situation, the problem lies in the backend storage, so no other protocol will perform better. However, P2P transferring will greatly reduce workload on the backend storage, so as to increase responsiveness. It's not 100% true, in my case at last. We fixed this problem by network interface driver, it causes kernel panic and readonly issues under heavy networking workload actually. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. Nova's image caching is file level, while VMThunder's is block-level. And VMThunder is for working in conjunction with Cinder, not Glance. VMThunder currently uses facebook's flashcache to realize caching, and dm-cache, bcache are also options in the future. Hm if you say bcache, dm-cache and flashcache, I'm just thinking if them could be leveraged by operation/best-practice level. btw, we are doing some works to make Glance to integrate Cinder as a unified block storage backend. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. Yes, on-demand transferring is what you mean by zero-copy, and caching is something close to CoR. In fact, we are working on a kernel module called foolcache that realize a true CoR. See https://github.com/lihuiba/dm-foolcache. Yup. And it's really interesting to me, will take a look, thanks for sharing. National Key Laboratory for Parallel and Distributed Processing, College of Computer Science, National University of Defense Technology, Changsha, Hunan Province, P.R. China 410073 At 2014-04-17 17:11:48,Zhi Yan Liu lzy@gmail.com wrote: On Thu, Apr 17, 2014 at 4:41 PM, lihuiba magazine.lihu...@163.com wrote: IMHO, zero-copy approach is better VMThunder's on-demand transferring is the same thing as your zero-copy
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
IMHO, zero-copy approach is better VMThunder's on-demand transferring is the same thing as your zero-copy approach. VMThunder is uses iSCSI as the transferring protocol, which is option #b of yours. Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. suppose booting one instance requires reading 300MB of data, so 500 ones require 150GB. Each of the storage server needs to send data at a rate of 150GB/2/60 = 1.25GB/s on average. This is absolutely a heavy burden even for high-end storage appliances. In production systems, this request (booting 500 VMs in one shot) will significantly disturb other running instances accessing the same storage nodes. VMThunder eliminates this problem by P2P transferring and on-compute-node caching. Even a pc server with one 1gb NIC (this is a true pc server!) can boot 500 VMs in a minute with ease. For the first time, VMThunder makes bulk provisioning of VMs practical for production cloud systems. This is the essential value of VMThunder. === From: Zhi Yan Liu lzy@gmail.com Date: 2014-04-17 0:02 GMT+08:00 Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org Hello Yongquan Fu, My thoughts: 1. Currently Nova has already supported image caching mechanism. It could caches the image on compute host which VM had provisioning from it before, and next provisioning (boot same image) doesn't need to transfer it again only if cache-manger clear it up. 2. P2P transferring and prefacing is something that still based on copy mechanism, IMHO, zero-copy approach is better, even transferring/prefacing could be optimized by such approach. (I have not check on-demand transferring of VMThunder, but it is a kind of transferring as well, at last from its literal meaning). And btw, IMO, we have two ways can go follow zero-copy idea: a. when Nova and Glance use same backend storage, we could use storage special CoW/snapshot approach to prepare VM disk instead of copy/transferring image bits (through HTTP/network or local copy). b. without unified storage, we could attach volume/LUN to compute node from backend storage as a base image, then do such CoW/snapshot on it to prepare root/ephemeral disk of VM. This way just like boot-from-volume but different is that we do CoW/snapshot on Nova side instead of Cinder/storage side. For option #a, we have already got some progress: https://blueprints.launchpad.net/nova/+spec/image-multiple-location https://blueprints.launchpad.net/nova/+spec/rbd-clone-image-handler https://blueprints.launchpad.net/nova/+spec/vmware-clone-image-handler Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. For vmThunder topic I think it sounds a good idea, IMO P2P, prefacing is one of optimized approach for image transferring valuably. zhiyan On Wed, Apr 16, 2014 at 9:14 PM, yongquan Fu quanyo...@gmail.com wrote: Dear all, We would like to present an extension to the vm-booting functionality of Nova when a number of homogeneous vms need to be launched at the same time. The motivation for our work is to increase the speed of provisioning vms for large-scale scientific computing and big data processing. In that case, we often need to boot tens and hundreds virtual machine instances at the same time. Currently, under the Openstack, we found that creating a large number of virtual machine instances is very time-consuming. The reason is the booting procedure is a centralized operation that involve performance bottlenecks. Before a virtual machine can be actually started, OpenStack either copy the image file (swift) or attach the image volume (cinder) from storage server to compute node via network. Booting a single VM need to read a large amount of image data from the image storage server. So creating a large number of virtual machine instances would cause a significant workload on the servers. The servers become quite busy even unavailable during the deployment phase. It would consume a very long time before the whole virtual machine cluster useable. Our extension is based on our work on vmThunder, a novel mechanism accelerating the deployment of large number virtual machine instances. It is written in Python, can be integrated with OpenStack easily. VMThunder addresses the problem described above by following improvements: on-demand
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
On Thu, Apr 17, 2014 at 4:41 PM, lihuiba magazine.lihu...@163.com wrote: IMHO, zero-copy approach is better VMThunder's on-demand transferring is the same thing as your zero-copy approach. VMThunder is uses iSCSI as the transferring protocol, which is option #b of yours. IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. suppose booting one instance requires reading 300MB of data, so 500 ones require 150GB. Each of the storage server needs to send data at a rate of 150GB/2/60 = 1.25GB/s on average. This is absolutely a heavy burden even for high-end storage appliances. In production systems, this request (booting 500 VMs in one shot) will significantly disturb other running instances accessing the same storage nodes. VMThunder eliminates this problem by P2P transferring and on-compute-node caching. Even a pc server with one 1gb NIC (this is a true pc server!) can boot 500 VMs in a minute with ease. For the first time, VMThunder makes bulk provisioning of VMs practical for production cloud systems. This is the essential value of VMThunder. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. zhiyan === From: Zhi Yan Liu lzy@gmail.com Date: 2014-04-17 0:02 GMT+08:00 Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org Hello Yongquan Fu, My thoughts: 1. Currently Nova has already supported image caching mechanism. It could caches the image on compute host which VM had provisioning from it before, and next provisioning (boot same image) doesn't need to transfer it again only if cache-manger clear it up. 2. P2P transferring and prefacing is something that still based on copy mechanism, IMHO, zero-copy approach is better, even transferring/prefacing could be optimized by such approach. (I have not check on-demand transferring of VMThunder, but it is a kind of transferring as well, at last from its literal meaning). And btw, IMO, we have two ways can go follow zero-copy idea: a. when Nova and Glance use same backend storage, we could use storage special CoW/snapshot approach to prepare VM disk instead of copy/transferring image bits (through HTTP/network or local copy). b. without unified storage, we could attach volume/LUN to compute node from backend storage as a base image, then do such CoW/snapshot on it to prepare root/ephemeral disk of VM. This way just like boot-from-volume but different is that we do CoW/snapshot on Nova side instead of Cinder/storage side. For option #a, we have already got some progress: https://blueprints.launchpad.net/nova/+spec/image-multiple-location https://blueprints.launchpad.net/nova/+spec/rbd-clone-image-handler https://blueprints.launchpad.net/nova/+spec/vmware-clone-image-handler Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. For vmThunder topic I think it sounds a good idea, IMO P2P, prefacing is one of optimized approach for image transferring valuably. zhiyan On Wed, Apr 16, 2014 at 9:14 PM, yongquan Fu quanyo...@gmail.com wrote: Dear all, We would like to present an extension to the vm-booting functionality of Nova when a number of homogeneous vms need to be launched at the same time. The motivation for our work is to increase the speed of provisioning vms for large-scale scientific computing and big data processing. In that case, we often need to boot tens and hundreds virtual machine instances at the same time. Currently, under the Openstack, we found that creating a large number of virtual machine instances is very
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
On 17 April 2014 11:11, Zhi Yan Liu lzy@gmail.com wrote: As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. This whole discussion reminded me of this: https://blueprints.launchpad.net/glance/+spec/glance-bittorrent-delivery http://tropicaldevel.wordpress.com/2013/01/11/an-image-transfers-service-for-openstack/ The general idea was that Glance would be able to serve images through torrents, enabling the capability for compute hosts to participate in image delivery. Well, the second part was where I thought it was going - I'm not sure if that was the intention. It didn't seem to go anywhere, but I thought it was a nifty idea. ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic.Yes, in this situation, the problem lies in the backend storage, so no otherprotocol will perform better. However, P2P transferring will greatly reduceworkload on the backend storage, so as to increase responsiveness. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. Nova's image caching is file level, while VMThunder's is block-level. And VMThunder is for working in conjunction with Cinder, not Glance. VMThunder currently uses facebook's flashcache to realize caching, and dm-cache, bcache are also options in the future. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. Yes, on-demand transferring is what you mean by zero-copy, and caching is something close to CoR. In fact, we are working on a kernel module called foolcache that realize a true CoR. See https://github.com/lihuiba/dm-foolcache. National Key Laboratory for Parallel and Distributed Processing, College of Computer Science, National University of Defense Technology, Changsha, Hunan Province, P.R. China 410073 At 2014-04-17 17:11:48,Zhi Yan Liu lzy@gmail.com wrote: On Thu, Apr 17, 2014 at 4:41 PM, lihuiba magazine.lihu...@163.com wrote: IMHO, zero-copy approach is better VMThunder's on-demand transferring is the same thing as your zero-copy approach. VMThunder is uses iSCSI as the transferring protocol, which is option #b of yours. IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. suppose booting one instance requires reading 300MB of data, so 500 ones require 150GB. Each of the storage server needs to send data at a rate of 150GB/2/60 = 1.25GB/s on average. This is absolutely a heavy burden even for high-end storage appliances. In production systems, this request (booting 500 VMs in one shot) will significantly disturb other running instances accessing the same storage nodes. VMThunder eliminates this problem by P2P transferring and on-compute-node caching. Even a pc server with one 1gb NIC (this is a true pc server!) can boot 500 VMs in a minute with ease. For the first time, VMThunder makes bulk provisioning of VMs practical for production cloud systems. This is the essential value of VMThunder. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. zhiyan === From: Zhi Yan Liu lzy@gmail.com Date: 2014-04-17 0:02 GMT+08:00 Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder To: OpenStack Development Mailing List (not for usage questions) openstack-dev@lists.openstack.org Hello Yongquan Fu, My thoughts: 1. Currently Nova has already supported image caching mechanism. It could caches the image on compute host which VM had provisioning from it before, and next provisioning (boot same image) doesn't need to transfer it again only if cache-manger clear it up. 2. P2P transferring and prefacing is something that still based on copy mechanism, IMHO, zero-copy approach is better, even transferring/prefacing could be optimized by such approach. (I have not check on-demand transferring of VMThunder, but it is a kind of transferring as well, at last from its literal meaning). And btw, IMO, we have two ways can go follow zero-copy idea: a. when Nova
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
glance-bittorrent-delivery and VMThunder have similar goals fast provisioning of large amount of VMs, and they share some ideas like P2P transferring, but they go with different techniques. VMThunder only downloads data blocks that are really used by VMs, so as to reduce bandwith and time required to provision. We have experiments showing that only a few hundred MB of data is needed to boot an mainstream OS like CentOS 6.x, Ubuntu 12.04, Windows 2008, etc., while the images are GBs or even tens of GBs large. National Key Laboratory for Parallel and Distributed Processing, College of Computer Science, National University of Defense Technology, Changsha, Hunan Province, P.R. China 410073 在 2014-04-17 19:06:27,Jesse Pretorius jesse.pretor...@gmail.com 写道: This whole discussion reminded me of this: https://blueprints.launchpad.net/glance/+spec/glance-bittorrent-delivery http://tropicaldevel.wordpress.com/2013/01/11/an-image-transfers-service-for-openstack/ The general idea was that Glance would be able to serve images through torrents, enabling the capability for compute hosts to participate in image delivery. Well, the second part was where I thought it was going - I'm not sure if that was the intention. It didn't seem to go anywhere, but I thought it was a nifty idea.___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
If you'd like to have a go at implementing this in nova's Juno release, then you need to create a new-style blueprint in the nova-specs repository. You can find more details about that process at https://wiki.openstack.org/wiki/Blueprints#Nova Some initial thoughts though, some of which have already been brought up: - _some_ libvirt drivers already have image caching. I am unsure if all of them do, I'd have to check. - we already have blueprints for better support of glance multiple image locations, it might be better to extend that work than to do something completely separate. - the xen driver already does bittorrent image delivery IIRC, you could take a look at how that do that. - pre-caching images has been proposed for libvirt for a long time, but never implemented. I think that's definitely something of interest to deployers. Cheers, Michael On Wed, Apr 16, 2014 at 11:14 PM, yongquan Fu quanyo...@gmail.com wrote: Dear all, We would like to present an extension to the vm-booting functionality of Nova when a number of homogeneous vms need to be launched at the same time. The motivation for our work is to increase the speed of provisioning vms for large-scale scientific computing and big data processing. In that case, we often need to boot tens and hundreds virtual machine instances at the same time. Currently, under the Openstack, we found that creating a large number of virtual machine instances is very time-consuming. The reason is the booting procedure is a centralized operation that involve performance bottlenecks. Before a virtual machine can be actually started, OpenStack either copy the image file (swift) or attach the image volume (cinder) from storage server to compute node via network. Booting a single VM need to read a large amount of image data from the image storage server. So creating a large number of virtual machine instances would cause a significant workload on the servers. The servers become quite busy even unavailable during the deployment phase. It would consume a very long time before the whole virtual machine cluster useable. Our extension is based on our work on vmThunder, a novel mechanism accelerating the deployment of large number virtual machine instances. It is written in Python, can be integrated with OpenStack easily. VMThunder addresses the problem described above by following improvements: on-demand transferring (network attached storage), compute node caching, P2P transferring and prefetching. VMThunder is a scalable and cost-effective accelerator for bulk provisioning of virtual machines. We hope to receive your feedbacks. Any comments are extremely welcome. Thanks in advance. PS: VMThunder enhanced nova blueprint: https://blueprints.launchpad.net/nova/+spec/thunderboost VMThunder standalone project: https://launchpad.net/vmthunder; VMThunder prototype: https://github.com/lihuiba/VMThunder VMThunder etherpad: https://etherpad.openstack.org/p/vmThunder VMThunder portal: http://www.vmthunder.org/ VMThunder paper: http://www.computer.org/csdl/trans/td/preprint/06719385.pdf Regards vmThunder development group PDL National University of Defense Technology ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev -- Rackspace Australia ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Replied as inline comments. On Thu, Apr 17, 2014 at 9:33 PM, lihuiba magazine.lihu...@163.com wrote: IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Yes, in this situation, the problem lies in the backend storage, so no other protocol will perform better. However, P2P transferring will greatly reduce workload on the backend storage, so as to increase responsiveness. It's not 100% true, in my case at last. We fixed this problem by network interface driver, it causes kernel panic and readonly issues under heavy networking workload actually. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. Nova's image caching is file level, while VMThunder's is block-level. And VMThunder is for working in conjunction with Cinder, not Glance. VMThunder currently uses facebook's flashcache to realize caching, and dm-cache, bcache are also options in the future. Hm if you say bcache, dm-cache and flashcache, I'm just thinking if them could be leveraged by operation/best-practice level. btw, we are doing some works to make Glance to integrate Cinder as a unified block storage backend. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. Yes, on-demand transferring is what you mean by zero-copy, and caching is something close to CoR. In fact, we are working on a kernel module called foolcache that realize a true CoR. See https://github.com/lihuiba/dm-foolcache. Yup. And it's really interesting to me, will take a look, thanks for sharing. National Key Laboratory for Parallel and Distributed Processing, College of Computer Science, National University of Defense Technology, Changsha, Hunan Province, P.R. China 410073 At 2014-04-17 17:11:48,Zhi Yan Liu lzy@gmail.com wrote: On Thu, Apr 17, 2014 at 4:41 PM, lihuiba magazine.lihu...@163.com wrote: IMHO, zero-copy approach is better VMThunder's on-demand transferring is the same thing as your zero-copy approach. VMThunder is uses iSCSI as the transferring protocol, which is option #b of yours. IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. suppose booting one instance requires reading 300MB of data, so 500 ones require 150GB. Each of the storage server needs to send data at a rate of 150GB/2/60 = 1.25GB/s on average. This is absolutely a heavy burden even for high-end storage appliances. In production systems, this request (booting 500 VMs in one shot) will significantly disturb other running instances accessing the same storage nodes. btw, I believe the case/numbers is not true as well, since remote image bits could be loaded on-demand instead of load them all on boot stage. zhiyan VMThunder eliminates this problem by P2P transferring and on-compute-node caching. Even a pc server with one 1gb NIC (this is a true pc server!) can boot 500 VMs in a minute with ease. For the first time, VMThunder makes bulk provisioning of VMs practical for production cloud systems. This is the essential value of VMThunder. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. zhiyan === From: Zhi Yan Liu lzy@gmail.com Date: 2014-04-17 0:02 GMT+08:00 Subject: Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder To: OpenStack Development
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
On Fri, Apr 18, 2014 at 5:19 AM, Michael Still mi...@stillhq.com wrote: If you'd like to have a go at implementing this in nova's Juno release, then you need to create a new-style blueprint in the nova-specs repository. You can find more details about that process at https://wiki.openstack.org/wiki/Blueprints#Nova Some initial thoughts though, some of which have already been brought up: - _some_ libvirt drivers already have image caching. I am unsure if all of them do, I'd have to check. Thanks for clarification. - we already have blueprints for better support of glance multiple image locations, it might be better to extend that work than to do something completely separate. Totally agreed. And I think currently seems there are two places (at least) could be leveraged: 1. Making this as an image download plug-ins for Nova, to be built-in or independent. I prefer to go this way, but need to make sure its context is enough for your case. 2. Making this as a built-in or independent image handler plug-ins, as a part of supporting of multiple-image-locations (on going) as Michael mentions here. zhiyan - the xen driver already does bittorrent image delivery IIRC, you could take a look at how that do that. - pre-caching images has been proposed for libvirt for a long time, but never implemented. I think that's definitely something of interest to deployers. Cheers, Michael On Wed, Apr 16, 2014 at 11:14 PM, yongquan Fu quanyo...@gmail.com wrote: Dear all, We would like to present an extension to the vm-booting functionality of Nova when a number of homogeneous vms need to be launched at the same time. The motivation for our work is to increase the speed of provisioning vms for large-scale scientific computing and big data processing. In that case, we often need to boot tens and hundreds virtual machine instances at the same time. Currently, under the Openstack, we found that creating a large number of virtual machine instances is very time-consuming. The reason is the booting procedure is a centralized operation that involve performance bottlenecks. Before a virtual machine can be actually started, OpenStack either copy the image file (swift) or attach the image volume (cinder) from storage server to compute node via network. Booting a single VM need to read a large amount of image data from the image storage server. So creating a large number of virtual machine instances would cause a significant workload on the servers. The servers become quite busy even unavailable during the deployment phase. It would consume a very long time before the whole virtual machine cluster useable. Our extension is based on our work on vmThunder, a novel mechanism accelerating the deployment of large number virtual machine instances. It is written in Python, can be integrated with OpenStack easily. VMThunder addresses the problem described above by following improvements: on-demand transferring (network attached storage), compute node caching, P2P transferring and prefetching. VMThunder is a scalable and cost-effective accelerator for bulk provisioning of virtual machines. We hope to receive your feedbacks. Any comments are extremely welcome. Thanks in advance. PS: VMThunder enhanced nova blueprint: https://blueprints.launchpad.net/nova/+spec/thunderboost VMThunder standalone project: https://launchpad.net/vmthunder; VMThunder prototype: https://github.com/lihuiba/VMThunder VMThunder etherpad: https://etherpad.openstack.org/p/vmThunder VMThunder portal: http://www.vmthunder.org/ VMThunder paper: http://www.computer.org/csdl/trans/td/preprint/06719385.pdf Regards vmThunder development group PDL National University of Defense Technology ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev -- Rackspace Australia ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
It's not 100% true, in my case at last. We fixed this problem by network interface driver, it causes kernel panic and readonly issues under heavy networking workload actually. Network traffic control could help. The point is to ensure no instance is starved to death. Traffic control can be done with tc. btw, we are doing some works to make Glance to integrate Cinder as a unified block storage backend. That sounds interesting. Is there some more materials? At 2014-04-18 06:05:23,Zhi Yan Liu lzy@gmail.com wrote: Replied as inline comments. On Thu, Apr 17, 2014 at 9:33 PM, lihuiba magazine.lihu...@163.com wrote: IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Yes, in this situation, the problem lies in the backend storage, so no other protocol will perform better. However, P2P transferring will greatly reduce workload on the backend storage, so as to increase responsiveness. It's not 100% true, in my case at last. We fixed this problem by network interface driver, it causes kernel panic and readonly issues under heavy networking workload actually. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. Nova's image caching is file level, while VMThunder's is block-level. And VMThunder is for working in conjunction with Cinder, not Glance. VMThunder currently uses facebook's flashcache to realize caching, and dm-cache, bcache are also options in the future. Hm if you say bcache, dm-cache and flashcache, I'm just thinking if them could be leveraged by operation/best-practice level. btw, we are doing some works to make Glance to integrate Cinder as a unified block storage backend. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. Yes, on-demand transferring is what you mean by zero-copy, and caching is something close to CoR. In fact, we are working on a kernel module called foolcache that realize a true CoR. See https://github.com/lihuiba/dm-foolcache. Yup. And it's really interesting to me, will take a look, thanks for sharing. National Key Laboratory for Parallel and Distributed Processing, College of Computer Science, National University of Defense Technology, Changsha, Hunan Province, P.R. China 410073 At 2014-04-17 17:11:48,Zhi Yan Liu lzy@gmail.com wrote: On Thu, Apr 17, 2014 at 4:41 PM, lihuiba magazine.lihu...@163.com wrote: IMHO, zero-copy approach is better VMThunder's on-demand transferring is the same thing as your zero-copy approach. VMThunder is uses iSCSI as the transferring protocol, which is option #b of yours. IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. suppose booting one instance requires reading 300MB of data, so 500 ones require 150GB. Each of the storage server needs to send data at a rate of 150GB/2/60 = 1.25GB/s on average. This is absolutely a heavy burden even for high-end storage appliances. In production systems, this request (booting 500 VMs in one shot) will significantly disturb other running instances accessing the same storage nodes. btw, I believe the case/numbers is not true as well, since remote image bits could be loaded on-demand instead of load them all on boot stage. zhiyan VMThunder eliminates this problem by P2P transferring and on-compute-node caching. Even a pc server with one 1gb NIC (this is a true pc server!) can boot 500 VMs in a minute with ease. For the first time, VMThunder makes bulk provisioning of VMs practical for production cloud systems. This is the essential value of VMThunder. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
- _some_ libvirt drivers already have image caching. I am unsure if all of them do, I'd have to check. '$instances_path/_base' is used to cache images downloaded fromglance, in file-level. While VMThunder employs find-grained block-level cacheing, for volumes served by cinder. - we already have blueprints for better support of glance multiple image locations, it might be better to extend that work than to do something completely separate. Is there a cinder multiple volume locations? We are considering to support something like that. - the xen driver already does bittorrent image delivery IIRC, you could take a look at how that do that. We are trying to do bittorrent image delivery for libvirt, too. - pre-caching images has been proposed for libvirt for a long time, but never implemented. I think that's definitely something of interest to deployers. What is pre-caching? Deploying images to compute nodes before they are used? Huiba Li National Key Laboratory for Parallel and Distributed Processing, College of Computer Science, National University of Defense Technology, Changsha, Hunan Province, P.R. China 410073 At 2014-04-18 05:19:23,Michael Still mi...@stillhq.com wrote: If you'd like to have a go at implementing this in nova's Juno release, then you need to create a new-style blueprint in the nova-specs repository. You can find more details about that process at https://wiki.openstack.org/wiki/Blueprints#Nova Some initial thoughts though, some of which have already been brought up: - _some_ libvirt drivers already have image caching. I am unsure if all of them do, I'd have to check. - we already have blueprints for better support of glance multiple image locations, it might be better to extend that work than to do something completely separate. - the xen driver already does bittorrent image delivery IIRC, you could take a look at how that do that. - pre-caching images has been proposed for libvirt for a long time, but never implemented. I think that's definitely something of interest to deployers. Cheers, Michael On Wed, Apr 16, 2014 at 11:14 PM, yongquan Fu quanyo...@gmail.com wrote: Dear all, We would like to present an extension to the vm-booting functionality of Nova when a number of homogeneous vms need to be launched at the same time. The motivation for our work is to increase the speed of provisioning vms for large-scale scientific computing and big data processing. In that case, we often need to boot tens and hundreds virtual machine instances at the same time. Currently, under the Openstack, we found that creating a large number of virtual machine instances is very time-consuming. The reason is the booting procedure is a centralized operation that involve performance bottlenecks. Before a virtual machine can be actually started, OpenStack either copy the image file (swift) or attach the image volume (cinder) from storage server to compute node via network. Booting a single VM need to read a large amount of image data from the image storage server. So creating a large number of virtual machine instances would cause a significant workload on the servers. The servers become quite busy even unavailable during the deployment phase. It would consume a very long time before the whole virtual machine cluster useable. Our extension is based on our work on vmThunder, a novel mechanism accelerating the deployment of large number virtual machine instances. It is written in Python, can be integrated with OpenStack easily. VMThunder addresses the problem described above by following improvements: on-demand transferring (network attached storage), compute node caching, P2P transferring and prefetching. VMThunder is a scalable and cost-effective accelerator for bulk provisioning of virtual machines. We hope to receive your feedbacks. Any comments are extremely welcome. Thanks in advance. PS: VMThunder enhanced nova blueprint: https://blueprints.launchpad.net/nova/+spec/thunderboost VMThunder standalone project: https://launchpad.net/vmthunder; VMThunder prototype: https://github.com/lihuiba/VMThunder VMThunder etherpad: https://etherpad.openstack.org/p/vmThunder VMThunder portal: http://www.vmthunder.org/ VMThunder paper: http://www.computer.org/csdl/trans/td/preprint/06719385.pdf Regards vmThunder development group PDL National University of Defense Technology ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev -- Rackspace Australia ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
On Fri, Apr 18, 2014 at 10:53 AM, lihuiba magazine.lihu...@163.com wrote: It's not 100% true, in my case at last. We fixed this problem by network interface driver, it causes kernel panic and readonly issues under heavy networking workload actually. Network traffic control could help. The point is to ensure no instance is starved to death. Traffic control can be done with tc. btw, I see but at the moment we had fixed it by network interface device driver instead of workaround - to limit network traffic slow down. btw, we are doing some works to make Glance to integrate Cinder as a unified block storage backend. That sounds interesting. Is there some more materials? There are few works done in Glance (https://blueprints.launchpad.net/glance/+spec/glance-cinder-driver ), but some work still need to be taken I'm sure. There are something on drafting, and some dependencies need to be resolved as well. At 2014-04-18 06:05:23,Zhi Yan Liu lzy@gmail.com wrote: Replied as inline comments. On Thu, Apr 17, 2014 at 9:33 PM, lihuiba magazine.lihu...@163.com wrote: IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Yes, in this situation, the problem lies in the backend storage, so no other protocol will perform better. However, P2P transferring will greatly reduce workload on the backend storage, so as to increase responsiveness. It's not 100% true, in my case at last. We fixed this problem by network interface driver, it causes kernel panic and readonly issues under heavy networking workload actually. As I said currently Nova already has image caching mechanism, so in this case P2P is just an approach could be used for downloading or preheating for image caching. Nova's image caching is file level, while VMThunder's is block-level. And VMThunder is for working in conjunction with Cinder, not Glance. VMThunder currently uses facebook's flashcache to realize caching, and dm-cache, bcache are also options in the future. Hm if you say bcache, dm-cache and flashcache, I'm just thinking if them could be leveraged by operation/best-practice level. btw, we are doing some works to make Glance to integrate Cinder as a unified block storage backend. I think P2P transferring/pre-caching sounds a good way to go, as I mentioned as well, but actually for the area I'd like to see something like zero-copy + CoR. On one hand we can leverage the capability of on-demand downloading image bits by zero-copy approach, on the other hand we can prevent to reading data from remote image every time by CoR. Yes, on-demand transferring is what you mean by zero-copy, and caching is something close to CoR. In fact, we are working on a kernel module called foolcache that realize a true CoR. See https://github.com/lihuiba/dm-foolcache. Yup. And it's really interesting to me, will take a look, thanks for sharing. National Key Laboratory for Parallel and Distributed Processing, College of Computer Science, National University of Defense Technology, Changsha, Hunan Province, P.R. China 410073 At 2014-04-17 17:11:48,Zhi Yan Liu lzy@gmail.com wrote: On Thu, Apr 17, 2014 at 4:41 PM, lihuiba magazine.lihu...@163.com wrote: IMHO, zero-copy approach is better VMThunder's on-demand transferring is the same thing as your zero-copy approach. VMThunder is uses iSCSI as the transferring protocol, which is option #b of yours. IMO we'd better to use backend storage optimized approach to access remote image from compute node instead of using iSCSI only. And from my experience, I'm sure iSCSI is short of stability under heavy I/O workload in product environment, it could causes either VM filesystem to be marked as readonly or VM kernel panic. Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. suppose booting one instance requires reading 300MB of data, so 500 ones require 150GB. Each of the storage server needs to send data at a rate of 150GB/2/60 = 1.25GB/s on average. This is absolutely a heavy burden even for high-end storage appliances. In production systems, this request (booting 500 VMs in one shot) will significantly disturb other running instances accessing the same storage nodes. btw, I believe the case/numbers is not true as well, since remote image bits could be loaded on-demand instead of load them all on boot stage. zhiyan VMThunder eliminates this problem by P2P transferring and on-compute-node caching. Even a pc server with one 1gb NIC (this is a true
Re: [openstack-dev] [Nova][blueprint] Accelerate the booting process of a number of vms via VMThunder
Hello Yongquan Fu, My thoughts: 1. Currently Nova has already supported image caching mechanism. It could caches the image on compute host which VM had provisioning from it before, and next provisioning (boot same image) doesn't need to transfer it again only if cache-manger clear it up. 2. P2P transferring and prefacing is something that still based on copy mechanism, IMHO, zero-copy approach is better, even transferring/prefacing could be optimized by such approach. (I have not check on-demand transferring of VMThunder, but it is a kind of transferring as well, at last from its literal meaning). And btw, IMO, we have two ways can go follow zero-copy idea: a. when Nova and Glance use same backend storage, we could use storage special CoW/snapshot approach to prepare VM disk instead of copy/transferring image bits (through HTTP/network or local copy). b. without unified storage, we could attach volume/LUN to compute node from backend storage as a base image, then do such CoW/snapshot on it to prepare root/ephemeral disk of VM. This way just like boot-from-volume but different is that we do CoW/snapshot on Nova side instead of Cinder/storage side. For option #a, we have already got some progress: https://blueprints.launchpad.net/nova/+spec/image-multiple-location https://blueprints.launchpad.net/nova/+spec/rbd-clone-image-handler https://blueprints.launchpad.net/nova/+spec/vmware-clone-image-handler Under #b approach, my former experience from our previous similar Cloud deployment (not OpenStack) was that: under 2 PC server storage nodes (general *local SAS disk*, without any storage backend) + 2-way/multi-path iSCSI + 1G network bandwidth, we can provisioning 500 VMs in a minute. For vmThunder topic I think it sounds a good idea, IMO P2P, prefacing is one of optimized approach for image transferring valuably. zhiyan On Wed, Apr 16, 2014 at 9:14 PM, yongquan Fu quanyo...@gmail.com wrote: Dear all, We would like to present an extension to the vm-booting functionality of Nova when a number of homogeneous vms need to be launched at the same time. The motivation for our work is to increase the speed of provisioning vms for large-scale scientific computing and big data processing. In that case, we often need to boot tens and hundreds virtual machine instances at the same time. Currently, under the Openstack, we found that creating a large number of virtual machine instances is very time-consuming. The reason is the booting procedure is a centralized operation that involve performance bottlenecks. Before a virtual machine can be actually started, OpenStack either copy the image file (swift) or attach the image volume (cinder) from storage server to compute node via network. Booting a single VM need to read a large amount of image data from the image storage server. So creating a large number of virtual machine instances would cause a significant workload on the servers. The servers become quite busy even unavailable during the deployment phase. It would consume a very long time before the whole virtual machine cluster useable. Our extension is based on our work on vmThunder, a novel mechanism accelerating the deployment of large number virtual machine instances. It is written in Python, can be integrated with OpenStack easily. VMThunder addresses the problem described above by following improvements: on-demand transferring (network attached storage), compute node caching, P2P transferring and prefetching. VMThunder is a scalable and cost-effective accelerator for bulk provisioning of virtual machines. We hope to receive your feedbacks. Any comments are extremely welcome. Thanks in advance. PS: VMThunder enhanced nova blueprint: https://blueprints.launchpad.net/nova/+spec/thunderboost VMThunder standalone project: https://launchpad.net/vmthunder; VMThunder prototype: https://github.com/lihuiba/VMThunder VMThunder etherpad: https://etherpad.openstack.org/p/vmThunder VMThunder portal: http://www.vmthunder.org/ VMThunder paper: http://www.computer.org/csdl/trans/td/preprint/06719385.pdf Regards vmThunder development group PDL National University of Defense Technology ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev ___ OpenStack-dev mailing list OpenStack-dev@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev