Re: [zfs-discuss] Comstar production-ready?
On Mar 4, 2009, at 7:04 AM, Jacob Ritorto wrote: Caution: I built a system like this and spent several weeks trying to get iscsi share working under Solaris 10 u6 and older. It would work fine for the first few hours but then performance would start to degrade, eventually becoming so poor as to actually cause panics on the iscsi initiator boxes. Couldn't find resolution through the various Solaris knowledge bases. Closest I got was to find out that there's a problem only in the *Solaris 10* iscsi target code that incorrectly frobs some counter when it shouldn't, violating the iscsi target specifications. The problem is fixed in Nevada/OpenSolaris. Long story short, I tried OpenSolaris 2008.11 and the iscsi crashes ceased and things ran smoothly. Not the solution I was hoping for, since this was to eventually be a prod box, but then Sun announced that I could purchase OpenSolaris support, so I was covered. On OS, my two big filers have been running really nicely for months and months now. Don't try to use Solaris 10 as a filer OS unless you can identify and resolve the iscsi target issue. The iSCSI Target Daemon in OpenSolaris 2008.xx, has been back ported to Solaris 10 u7. - Jim On Wed, Mar 4, 2009 at 2:47 AM, Scott Lawson scott.law...@manukau.ac.nz wrote: Stephen Nelson-Smith wrote: Hi, I recommended a ZFS-based archive solution to a client needing to have a network-based archive of 15TB of data in a remote datacentre. I based this on an X2200 + J4400, Solaris 10 + rsync. This was enthusiastically received, to the extent that the client is now requesting that their live system (15TB data on cheap SAN and Linux LVM) be replaced with a ZFS-based system. The catch is that they're not ready to move their production systems off Linux - so web, db and app layer will all still be on RHEL 5. At some point I am sure you will convince them to see the light! ;) As I see it, if they want to benefit from ZFS at the storage layer, the obvious solution would be a NAS system, such as a 7210, or something buillt from a JBOD and a head node that does something similar. The 7210 is out of budget - and I'm not quite sure how it presents its storage - is it NFS/CIFS? The 7000 series devices can present NFS, CIFS and iSCSI. Looks very nice if you need a nice Gui / Don't know command line / need nice analytics. I had a play with one the other day and am hoping to get my mit's on one shortly for testing. I would like to give it a real gd crack with VMWare for VDI VM's. If so, presumably it would be relatively easy to build something equivalent, but without the (awesome) interface. For sure the above gear would be fine for that. If you use standard Solaris 10 10/08 you have NFS and iSCSI ability directly in the OS and also available to be supported via a support contract if needed. Best bet would probably be NFS for the Linux machines, but you would need to test in *their* environment with *their* workload. The interesting alternative is to set up Comstar on SXCE, create zpools and volumes, and make these available either over a fibre infrastructure, or iSCSI. I'm quite excited by this as a solution, but I'm not sure if it's really production ready. If you want fibre channel target then you will need to use OpenSolaris or SXDE I believe. It's not available in mainstream Solaris yet. I am personally waiting till then when it has been *well* tested in the bleeding edge community. I have too much data to take big risks with it. What other options are there, and what advice/experience can you share? I do very similar stuff here with J4500's and T2K's for compliance archives, NFS and iSCSI targets for Windows machines. Works fine for me. Biggest system is 48TB on J4500 for Veritas Netbackup DDT staging volumes. Very good throughput indeed. Perfect in fact, based on the large files that are created in this environment. One of these J4500's can keep 4 LTO4 drives in a SL500 saturated with data on a T5220. (4 streams at ~160 MB/sec) I think you have pretty much the right idea though. Certainly if you use Sun kit you will be able to deliver a commercially supported solution for them. ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Re: [zfs-discuss] Comstar production-ready?
Caution: I built a system like this and spent several weeks trying to get iscsi share working under Solaris 10 u6 and older. It would work fine for the first few hours but then performance would start to degrade, eventually becoming so poor as to actually cause panics on the iscsi initiator boxes. Couldn't find resolution through the various Solaris knowledge bases. Closest I got was to find out that there's a problem only in the *Solaris 10* iscsi target code that incorrectly frobs some counter when it shouldn't, violating the iscsi target specifications. The problem is fixed in Nevada/OpenSolaris. Long story short, I tried OpenSolaris 2008.11 and the iscsi crashes ceased and things ran smoothly. Not the solution I was hoping for, since this was to eventually be a prod box, but then Sun announced that I could purchase OpenSolaris support, so I was covered. On OS, my two big filers have been running really nicely for months and months now. Don't try to use Solaris 10 as a filer OS unless you can identify and resolve the iscsi target issue. On Wed, Mar 4, 2009 at 2:47 AM, Scott Lawson scott.law...@manukau.ac.nz wrote: Stephen Nelson-Smith wrote: Hi, I recommended a ZFS-based archive solution to a client needing to have a network-based archive of 15TB of data in a remote datacentre. I based this on an X2200 + J4400, Solaris 10 + rsync. This was enthusiastically received, to the extent that the client is now requesting that their live system (15TB data on cheap SAN and Linux LVM) be replaced with a ZFS-based system. The catch is that they're not ready to move their production systems off Linux - so web, db and app layer will all still be on RHEL 5. At some point I am sure you will convince them to see the light! ;) As I see it, if they want to benefit from ZFS at the storage layer, the obvious solution would be a NAS system, such as a 7210, or something buillt from a JBOD and a head node that does something similar. The 7210 is out of budget - and I'm not quite sure how it presents its storage - is it NFS/CIFS? The 7000 series devices can present NFS, CIFS and iSCSI. Looks very nice if you need a nice Gui / Don't know command line / need nice analytics. I had a play with one the other day and am hoping to get my mit's on one shortly for testing. I would like to give it a real gd crack with VMWare for VDI VM's. If so, presumably it would be relatively easy to build something equivalent, but without the (awesome) interface. For sure the above gear would be fine for that. If you use standard Solaris 10 10/08 you have NFS and iSCSI ability directly in the OS and also available to be supported via a support contract if needed. Best bet would probably be NFS for the Linux machines, but you would need to test in *their* environment with *their* workload. The interesting alternative is to set up Comstar on SXCE, create zpools and volumes, and make these available either over a fibre infrastructure, or iSCSI. I'm quite excited by this as a solution, but I'm not sure if it's really production ready. If you want fibre channel target then you will need to use OpenSolaris or SXDE I believe. It's not available in mainstream Solaris yet. I am personally waiting till then when it has been *well* tested in the bleeding edge community. I have too much data to take big risks with it. What other options are there, and what advice/experience can you share? I do very similar stuff here with J4500's and T2K's for compliance archives, NFS and iSCSI targets for Windows machines. Works fine for me. Biggest system is 48TB on J4500 for Veritas Netbackup DDT staging volumes. Very good throughput indeed. Perfect in fact, based on the large files that are created in this environment. One of these J4500's can keep 4 LTO4 drives in a SL500 saturated with data on a T5220. (4 streams at ~160 MB/sec) I think you have pretty much the right idea though. Certainly if you use Sun kit you will be able to deliver a commercially supported solution for them. ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Re: [zfs-discuss] Comstar production-ready?
On 4-Mar-09, at 2:07 AM, Stephen Nelson-Smith wrote: Hi, I recommended a ZFS-based archive solution to a client needing to have a network-based archive of 15TB of data in a remote datacentre. I based this on an X2200 + J4400, Solaris 10 + rsync. This was enthusiastically received, to the extent that the client is now requesting that their live system (15TB data on cheap SAN and Linux LVM) be replaced with a ZFS-based system. The catch is that they're not ready to move their production systems off Linux - so web, db and app layer will all still be on RHEL 5. Considered zones/virtualisation? Just a wild thought. --Toby As I see it, if they want to benefit from ZFS at the storage layer, the obvious solution would be a NAS system, ... Thanks, S. -- Stephen Nelson-Smith Technical Director Atalanta Systems Ltd www.atalanta-systems.com ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Re: [zfs-discuss] Comstar production-ready?
On Wed, 4 Mar 2009, Stephen Nelson-Smith wrote: The interesting alternative is to set up Comstar on SXCE, create zpools and volumes, and make these available either over a fibre infrastructure, or iSCSI. I'm quite excited by this as a solution, but I'm not sure if it's really production ready. While this is indeed exciting, the solutions you have proposed vary considerably in the type of functionality they offer. Comstar and iSCSI provide access to a storage volume similar to SAN storage. This volume is then formatted with some alien filesystem which is unlikely to support the robust features of ZFS. Even though the storage volume is implemented in robust ZFS, the client still has the ability to scramble its own filesystem. ZFS snapshots can help defend against that by allowing to rewind the entire content of the storage volume to a former point in time. With the NFS/CIFS server model, only ZFS is used. There is no dependence on a client filesystem. With the Comstar/iSCSI approach, you are balkanizing (http://en.wikipedia.org/wiki/Balkanization) your storage so that each client owns its own filesystems without ability to share the data unless the client does it. With the native ZFS server approach, all clients share the pool storage and can share files on the server if the server allows it. A drawback of the native ZFS server approach is that the server needs to know about the users on the clients in order to support access control. Regardless, there are cases where Comstar/iSCSI makes the most sense, or the ZFS fileserver makes the most sense. Bob -- Bob Friesenhahn bfrie...@simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer,http://www.GraphicsMagick.org/ ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Re: [zfs-discuss] Comstar production-ready?
Jacob Ritorto wrote: Caution: I built a system like this and spent several weeks trying to get iscsi share working under Solaris 10 u6 and older. It would work fine for the first few hours but then performance would start to degrade, eventually becoming so poor as to actually cause panics on the iscsi initiator boxes. Couldn't find resolution through the various Solaris knowledge bases. Closest I got was to find out that there's a problem only in the *Solaris 10* iscsi target code that incorrectly frobs some counter when it shouldn't, violating the iscsi target specifications. The problem is fixed in Nevada/OpenSolaris. Can't say I have had a problem myself. Initiator is the default Microsoft Vista initiator. Mine has been running for at least 6-9 months fine. It doesn't get a absolute hammering though. Using it to provide extra storage to IT staff desktops who need it and at the same time allowing staff to play with iSCSI. I run a largish fibre channel shop and prefer that mostly anyway. Long story short, I tried OpenSolaris 2008.11 and the iscsi crashes ceased and things ran smoothly. Not the solution I was hoping for, since this was to eventually be a prod box, but then Sun announced that I could purchase OpenSolaris support, so I was covered. On OS, my two big filers have been running really nicely for months and months now. If you get Solaris 10 support sun will provide fixes for that too I imagine. Again, can't say I have had a problem myself. But as I mentioned in my previous email, I can't stress how important it is to *test* a solution in your environment with *your* workload with the hardware/OS *you* choose. The Sun try and buy on hardware is a great way to do this relatively risk free. If it doesn't work send it back. Also they have startup essentials which will potentially allow you to ge the try and buy hardware in the cheap if you are a new customer. Don't try to use Solaris 10 as a filer OS unless you can identify and resolve the iscsi target issue. If iSCSI is truly broken then one could log a support call on this is you take basic maintenance. This is cheaper the RHEL for the entry level stuff by the way... flame Being that this is a Linux shop you are selling into, OpenSolaris might be the best way to go as the GNU userland might be more familiar to them and they might not understand having to change their shell paths to get the userland that they want ;) /flame On Wed, Mar 4, 2009 at 2:47 AM, Scott Lawson scott.law...@manukau.ac.nz wrote: Stephen Nelson-Smith wrote: Hi, I recommended a ZFS-based archive solution to a client needing to have a network-based archive of 15TB of data in a remote datacentre. I based this on an X2200 + J4400, Solaris 10 + rsync. This was enthusiastically received, to the extent that the client is now requesting that their live system (15TB data on cheap SAN and Linux LVM) be replaced with a ZFS-based system. The catch is that they're not ready to move their production systems off Linux - so web, db and app layer will all still be on RHEL 5. At some point I am sure you will convince them to see the light! ;) As I see it, if they want to benefit from ZFS at the storage layer, the obvious solution would be a NAS system, such as a 7210, or something buillt from a JBOD and a head node that does something similar. The 7210 is out of budget - and I'm not quite sure how it presents its storage - is it NFS/CIFS? The 7000 series devices can present NFS, CIFS and iSCSI. Looks very nice if you need a nice Gui / Don't know command line / need nice analytics. I had a play with one the other day and am hoping to get my mit's on one shortly for testing. I would like to give it a real gd crack with VMWare for VDI VM's. If so, presumably it would be relatively easy to build something equivalent, but without the (awesome) interface. For sure the above gear would be fine for that. If you use standard Solaris 10 10/08 you have NFS and iSCSI ability directly in the OS and also available to be supported via a support contract if needed. Best bet would probably be NFS for the Linux machines, but you would need to test in *their* environment with *their* workload. The interesting alternative is to set up Comstar on SXCE, create zpools and volumes, and make these available either over a fibre infrastructure, or iSCSI. I'm quite excited by this as a solution, but I'm not sure if it's really production ready. If you want fibre channel target then you will need to use OpenSolaris or SXDE I believe. It's not available in mainstream Solaris yet. I am personally waiting till then when it has been *well* tested in the bleeding edge community. I have too much data to take big risks with it. What other options are there, and what advice/experience can you share? I do very similar stuff here with J4500's and T2K's for compliance archives, NFS and iSCSI targets for
Re: [zfs-discuss] Comstar production-ready?
Right on the money there Bob. Without knowing more detail about the clients workload, it would be hard to advise either way. I would imagine based purely on the small amount of info around the clients apps and workload that NFS would most likely be the appropriate solution on top of ZFS. You will make more efficient use of the your ZFS storage this way and provide all the niceties like snapshots and rollbacks from the Solaris based filer whilst still maintaining Linux front ends. Do take heed to the various list posts around ZFS/NFS with certain types of workloads however. Bob Friesenhahn wrote: On Wed, 4 Mar 2009, Stephen Nelson-Smith wrote: The interesting alternative is to set up Comstar on SXCE, create zpools and volumes, and make these available either over a fibre infrastructure, or iSCSI. I'm quite excited by this as a solution, but I'm not sure if it's really production ready. While this is indeed exciting, the solutions you have proposed vary considerably in the type of functionality they offer. Comstar and iSCSI provide access to a storage volume similar to SAN storage. This volume is then formatted with some alien filesystem which is unlikely to support the robust features of ZFS. Even though the storage volume is implemented in robust ZFS, the client still has the ability to scramble its own filesystem. ZFS snapshots can help defend against that by allowing to rewind the entire content of the storage volume to a former point in time. With the NFS/CIFS server model, only ZFS is used. There is no dependence on a client filesystem. With the Comstar/iSCSI approach, you are balkanizing (http://en.wikipedia.org/wiki/Balkanization) your storage so that each client owns its own filesystems without ability to share the data unless the client does it. With the native ZFS server approach, all clients share the pool storage and can share files on the server if the server allows it. A drawback of the native ZFS server approach is that the server needs to know about the users on the clients in order to support access control. Regardless, there are cases where Comstar/iSCSI makes the most sense, or the ZFS fileserver makes the most sense. Bob -- Bob Friesenhahn bfrie...@simple.dallas.tx.us, http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer,http://www.GraphicsMagick.org/ ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss -- ___ Scott Lawson Systems Architect Manukau Institute of Technology Information Communication Technology Services Private Bag 94006 Manukau City Auckland New Zealand Phone : +64 09 968 7611 Fax: +64 09 968 7641 Mobile : +64 27 568 7611 mailto:sc...@manukau.ac.nz http://www.manukau.ac.nz perl -e 'print $i=pack(c5,(41*2),sqrt(7056),(unpack(c,H)-2),oct(115),10);' ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Re: [zfs-discuss] Comstar production-ready?
On Wed, Mar 4, 2009 at 2:07 PM, Stephen Nelson-Smith sanel...@gmail.com wrote: As I see it, if they want to benefit from ZFS at the storage layer, the obvious solution would be a NAS system, such as a 7210, or something buillt from a JBOD and a head node that does something similar. The 7210 is out of budget - and I'm not quite sure how it presents its storage - is it NFS/CIFS? If so, presumably it would be it can also share block device (zvol) with iscsi The interesting alternative is to set up Comstar on SXCE, create zpools and volumes, and make these available either over a fibre infrastructure, or iSCSI. I'm quite excited by this as a solution, but I'm not sure if it's really production ready. If you want production rady software, starting from Solaris 10 8/07 Release you can create a ZFS volume as a Solaris iSCSI target device by setting the shareiscsi property on the ZFS volume. It's not Comstar, but it works. You may want consider opensolaris (I like it better than SXCE) instead of solaris if you want to stay on bleeding-edge, or even Nexenta which recentely integrated Comstar http://www.gulecha.org/2009/03/03/nexenta-iscsi-with-comstarzfs-integration/ Regards, Fajar ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Re: [zfs-discuss] Comstar production-ready?
Hi Stephen, NexentaStor v1.1.5+ could be an alternative, I think. And it includes new cool COMSTAR integration, i.e. ZFS shareiscsi property actually implements COMSTAR iSCSI target share functionality not available in SXCE. http://www.nexenta.com/nexentastor-relnotes On Wed, 2009-03-04 at 07:07 +, Stephen Nelson-Smith wrote: Hi, I recommended a ZFS-based archive solution to a client needing to have a network-based archive of 15TB of data in a remote datacentre. I based this on an X2200 + J4400, Solaris 10 + rsync. This was enthusiastically received, to the extent that the client is now requesting that their live system (15TB data on cheap SAN and Linux LVM) be replaced with a ZFS-based system. The catch is that they're not ready to move their production systems off Linux - so web, db and app layer will all still be on RHEL 5. As I see it, if they want to benefit from ZFS at the storage layer, the obvious solution would be a NAS system, such as a 7210, or something buillt from a JBOD and a head node that does something similar. The 7210 is out of budget - and I'm not quite sure how it presents its storage - is it NFS/CIFS? If so, presumably it would be relatively easy to build something equivalent, but without the (awesome) interface. The interesting alternative is to set up Comstar on SXCE, create zpools and volumes, and make these available either over a fibre infrastructure, or iSCSI. I'm quite excited by this as a solution, but I'm not sure if it's really production ready. What other options are there, and what advice/experience can you share? Thanks, S. ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Re: [zfs-discuss] Comstar production-ready?
Stephen Nelson-Smith wrote: Hi, I recommended a ZFS-based archive solution to a client needing to have a network-based archive of 15TB of data in a remote datacentre. I based this on an X2200 + J4400, Solaris 10 + rsync. This was enthusiastically received, to the extent that the client is now requesting that their live system (15TB data on cheap SAN and Linux LVM) be replaced with a ZFS-based system. The catch is that they're not ready to move their production systems off Linux - so web, db and app layer will all still be on RHEL 5. At some point I am sure you will convince them to see the light! ;) As I see it, if they want to benefit from ZFS at the storage layer, the obvious solution would be a NAS system, such as a 7210, or something buillt from a JBOD and a head node that does something similar. The 7210 is out of budget - and I'm not quite sure how it presents its storage - is it NFS/CIFS? The 7000 series devices can present NFS, CIFS and iSCSI. Looks very nice if you need a nice Gui / Don't know command line / need nice analytics. I had a play with one the other day and am hoping to get my mit's on one shortly for testing. I would like to give it a real gd crack with VMWare for VDI VM's. If so, presumably it would be relatively easy to build something equivalent, but without the (awesome) interface. For sure the above gear would be fine for that. If you use standard Solaris 10 10/08 you have NFS and iSCSI ability directly in the OS and also available to be supported via a support contract if needed. Best bet would probably be NFS for the Linux machines, but you would need to test in *their* environment with *their* workload. The interesting alternative is to set up Comstar on SXCE, create zpools and volumes, and make these available either over a fibre infrastructure, or iSCSI. I'm quite excited by this as a solution, but I'm not sure if it's really production ready. If you want fibre channel target then you will need to use OpenSolaris or SXDE I believe. It's not available in mainstream Solaris yet. I am personally waiting till then when it has been *well* tested in the bleeding edge community. I have too much data to take big risks with it. What other options are there, and what advice/experience can you share? I do very similar stuff here with J4500's and T2K's for compliance archives, NFS and iSCSI targets for Windows machines. Works fine for me. Biggest system is 48TB on J4500 for Veritas Netbackup DDT staging volumes. Very good throughput indeed. Perfect in fact, based on the large files that are created in this environment. One of these J4500's can keep 4 LTO4 drives in a SL500 saturated with data on a T5220. (4 streams at ~160 MB/sec) I think you have pretty much the right idea though. Certainly if you use Sun kit you will be able to deliver a commercially supported solution for them. Thanks, S. -- _ Scott Lawson Systems Architect Information Communication Technology Services Manukau Institute of Technology Private Bag 94006 South Auckland Mail Centre Manukau 2240 Auckland New Zealand Phone : +64 09 968 7611 Fax: +64 09 968 7641 Mobile : +64 27 568 7611 mailto:sc...@manukau.ac.nz http://www.manukau.ac.nz __ perl -e 'print $i=pack(c5,(41*2),sqrt(7056),(unpack(c,H)-2),oct(115),10);' __ ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
