Re: [Openstack-operators] Passing entire disks to instances
Hi steve, The blockdevicedriver looks like exactly what i need. I don't know how I missed it. Thank you! David On August 30, 2015 10:53:40 AM EDT, Steve Gordon wrote: >- Original Message - >> From: "David Arroyo" >> To: openstack-operators@lists.openstack.org >> >> Hello, >> >> I would like to pass entire disks to my openstack instances. I have >> some IO-bound workloads and would like to avoid any overhead or >> contention by giving an instance access to N additional disks, such >> that they do not share that disk with other guests on the same >compute >> node. >> >> These additional disks do not need to last longer than the instances >> themselves; they are "ephemeral" in that regard. There is no need for >> backup, no need for instance or data migration, and no need for >running >> the instance on a separate compute node from the extra disks. >> Effectively I want to have an "extra disks" property, as part of a >> flavor or image, that is handled like vcpus without overcommit. >> >> I have been doing some research but have not yet found an obvious way >> to do this in openstack. Does anyone else have a similar use case, >and >> how have you handled it? To be more specific, we run some very large >> Cassandra clusters on physical hardware, with excellent performance. >> Cassandra is largely IO-bound, and we want to virtualize it without >> introducing unnecessary IO latency. >> >> Cheers, >> David > >Have you looked at the BlockDeviceDriver as described here: > >http://cloudgeekz.com/71/how-to-setup-openstack-to-use-local-disks-for-instances.html > >-Steve -- Sent from my Android device with K-9 Mail. Please excuse my brevity.___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
Re: [Openstack-operators] Passing entire disks to instances
- Original Message - > From: "David Arroyo" > To: openstack-operators@lists.openstack.org > > Hello, > > I would like to pass entire disks to my openstack instances. I have > some IO-bound workloads and would like to avoid any overhead or > contention by giving an instance access to N additional disks, such > that they do not share that disk with other guests on the same compute > node. > > These additional disks do not need to last longer than the instances > themselves; they are "ephemeral" in that regard. There is no need for > backup, no need for instance or data migration, and no need for running > the instance on a separate compute node from the extra disks. > Effectively I want to have an "extra disks" property, as part of a > flavor or image, that is handled like vcpus without overcommit. > > I have been doing some research but have not yet found an obvious way > to do this in openstack. Does anyone else have a similar use case, and > how have you handled it? To be more specific, we run some very large > Cassandra clusters on physical hardware, with excellent performance. > Cassandra is largely IO-bound, and we want to virtualize it without > introducing unnecessary IO latency. > > Cheers, > David Have you looked at the BlockDeviceDriver as described here: http://cloudgeekz.com/71/how-to-setup-openstack-to-use-local-disks-for-instances.html -Steve ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
Re: [Openstack-operators] Passing entire disks to instances
Would an OpenStack cinder volume meet your needs ? Tim From: David Arroyo [mailto:dr...@aqwari.net] Sent: 29 August 2015 17:17 To: openstack-operators@lists.openstack.org Subject: [Openstack-operators] Passing entire disks to instances Hello, I would like to pass entire disks to my openstack instances. I have some IO-bound workloads and would like to avoid any overhead or contention by giving an instance access to N additional disks, such that they do not share that disk with other guests on the same compute node. These additional disks do not need to last longer than the instances themselves; they are "ephemeral" in that regard. There is no need for backup, no need for instance or data migration, and no need for running the instance on a separate compute node from the extra disks. Effectively I want to have an "extra disks" property, as part of a flavor or image, that is handled like vcpus without overcommit. I have been doing some research but have not yet found an obvious way to do this in openstack. Does anyone else have a similar use case, and how have you handled it? To be more specific, we run some very large Cassandra clusters on physical hardware, with excellent performance. Cassandra is largely IO-bound, and we want to virtualize it without introducing unnecessary IO latency. Cheers, David ___ OpenStack-operators mailing list OpenStack-operators@lists.openstack.org http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators