Re: [ovirt-devel] [RFE] treat local NFS storage as localfs
On 21 Dec 2016, at 16:26, Martin Sivak <msivak@redhat.com> wrote:
Hi,
Hope this get's in. This seems less overhead than a complete hyperconverged gluster setup.
But NFS still is a single point of failure. Hyperconverged is supposed to address that.
In order to improve performance, disk I/O bound VMs can be pinned to a host with local storage. However there still is a performance drawback of NFS layers. Treating a local NFS storage as a local storage improves performance for VMs pinned to host.
So VMs on one host will get better IO performance and the others will still use NFS as they do now.
It is an interesting idea, I am just not sure if having poor-man's hyperconverged setup with all the drawbacks of NFS is worth it. Imagine for example what happens when that storage provider host needs to be fenced or put into maintenance. The whole cluster would go down (all VMs would lose storage connection, not just the VMs from the affected host).
I will let someone from the storage team to respond to this, but I do not think that trading performance (each host has its own local storage) and resilience (well, at least one failing host does not affect the others) for migrations is a good deal.
If disk performance is critical then there is an option to use direct access on local host using either PCI passthrough of a local storage controller or SCSI passthrough of LUNs.
-- Martin Sivak SLA / oVirt
On Wed, Dec 21, 2016 at 2:18 PM, Sven Kieske <s.kieske@mittwald.de> wrote:
On 21/12/16 11:44, Pavel Gashev wrote: Hello,
I'd like to introduce a RFE that allows to use a local storage in multi server environments https://bugzilla.redhat.com/show_bug.cgi?id=1406412
Most servers have a local storage. Some servers have very reliable storages with hardware RAID controllers and battery units.
Example user cases: https://www.mail-archive.com/users@ovirt.org/msg36719.html https://www.mail-archive.com/users@ovirt.org/msg36772.html
The best way to use local storage in multi server "shared" datacenters is exporting it over NFS. Using NFS allows to move disks and VMs among servers.
In order to improve performance, disk I/O bound VMs can be pinned to a host with local storage. However there still is a performance drawback of NFS layers. Treating a local NFS storage as a local storage improves performance for VMs pinned to host.
Currently setting up of NFS exports is out of scope of oVirt. However this would be a way to get rid of "Local/Shared" storage types of datacenter. So that all storages are shared, but local storages are used as local.
Any questions/comments are welcome.
Specifically I'd like to request for comment on potential data integrity issues during online VM or disk migration between NFS and localfs.
Just let me say that I really like this as an end user.
Hope this get's in. This seems less overhead than a complete hyperconverged gluster setup.
-- Mit freundlichen Grüßen / Regards
Sven Kieske
Systemadministrator Mittwald CM Service GmbH & Co. KG Königsberger Straße 6 32339 Espelkamp T: +495772 293100 F: +495772 293333 https://www.mittwald.de Geschäftsführer: Robert Meyer St.Nr.: 331/5721/1033, USt-IdNr.: DE814773217, HRA 6640, AG Bad Oeynhausen Komplementärin: Robert Meyer Verwaltungs GmbH, HRB 13260, AG Bad Oeynhausen
_______________________________________________ Devel mailing list Devel@ovirt.org http://lists.ovirt.org/mailman/listinfo/devel
Devel mailing list Devel@ovirt.org http://lists.ovirt.org/mailman/listinfo/devel
Also NFS had locking issues on loopback. Not really possible to do this with NFS. What is the issue with using a Gluster local replicate 1 for this? Yaniv Dary Technical Product Manager Red Hat Israel Ltd. 34 Jerusalem Road Building A, 4th floor Ra'anana, Israel 4350109 Tel : +972 (9) 7692306 8272306 Email: ydary@redhat.com IRC : ydary On Wed, Dec 21, 2016 at 8:03 PM, Michal Skrivanek <mskrivan@redhat.com> wrote:
On 21 Dec 2016, at 16:26, Martin Sivak <msivak@redhat.com> wrote:
Hi,
Hope this get's in. This seems less overhead than a complete hyperconverged gluster setup.
But NFS still is a single point of failure. Hyperconverged is supposed to address that.
In order to improve performance, disk I/O bound VMs can be pinned to a host with local storage. However there still is a performance drawback of NFS layers. Treating a local NFS storage as a local storage improves performance for VMs pinned to host.
So VMs on one host will get better IO performance and the others will still use NFS as they do now.
It is an interesting idea, I am just not sure if having poor-man's hyperconverged setup with all the drawbacks of NFS is worth it. Imagine for example what happens when that storage provider host needs to be fenced or put into maintenance. The whole cluster would go down (all VMs would lose storage connection, not just the VMs from the affected host).
I will let someone from the storage team to respond to this, but I do not think that trading performance (each host has its own local storage) and resilience (well, at least one failing host does not affect the others) for migrations is a good deal.
If disk performance is critical then there is an option to use direct access on local host using either PCI passthrough of a local storage controller or SCSI passthrough of LUNs.
-- Martin Sivak SLA / oVirt
On Wed, Dec 21, 2016 at 2:18 PM, Sven Kieske <s.kieske@mittwald.de>
wrote:
On 21/12/16 11:44, Pavel Gashev wrote: Hello,
I'd like to introduce a RFE that allows to use a local storage in multi server environments https://bugzilla.redhat.com/ show_bug.cgi?id=1406412
Most servers have a local storage. Some servers have very reliable storages with hardware RAID controllers and battery units.
Example user cases: https://www.mail-archive.com/users@ovirt.org/msg36719.html https://www.mail-archive.com/users@ovirt.org/msg36772.html
The best way to use local storage in multi server "shared" datacenters is exporting it over NFS. Using NFS allows to move disks and VMs among servers.
In order to improve performance, disk I/O bound VMs can be pinned to a host with local storage. However there still is a performance drawback of NFS layers. Treating a local NFS storage as a local storage improves performance for VMs pinned to host.
Currently setting up of NFS exports is out of scope of oVirt. However this would be a way to get rid of "Local/Shared" storage types of datacenter. So that all storages are shared, but local storages are used as local.
Any questions/comments are welcome.
Specifically I'd like to request for comment on potential data integrity issues during online VM or disk migration between NFS and localfs.
Just let me say that I really like this as an end user.
Hope this get's in. This seems less overhead than a complete hyperconverged gluster setup.
-- Mit freundlichen Grüßen / Regards
Sven Kieske
Systemadministrator Mittwald CM Service GmbH & Co. KG Königsberger Straße 6 32339 Espelkamp T: +495772 293100 F: +495772 293333 https://www.mittwald.de Geschäftsführer: Robert Meyer St.Nr.: 331/5721/1033, USt-IdNr.: DE814773217, HRA 6640, AG Bad Oeynhausen Komplementärin: Robert Meyer Verwaltungs GmbH, HRB 13260, AG Bad Oeynhausen
_______________________________________________ Devel mailing list Devel@ovirt.org http://lists.ovirt.org/mailman/listinfo/devel
Devel mailing list Devel@ovirt.org http://lists.ovirt.org/mailman/listinfo/devel
_______________________________________________ Devel mailing list Devel@ovirt.org http://lists.ovirt.org/mailman/listinfo/devel
Michal, VMs with attached devices are bound to host. When I use NFS I can migrate my VM and disks to another host. On Wed, 2016-12-21 at 19:03 +0100, Michal Skrivanek wrote:
If disk performance is critical then there is an option to use direct access on local host using either PCI passthrough of a local storage controller or SCSI passthrough of LUNs.
On 22 Dec 2016, at 16:12, Pavel Gashev <Pax@acronis.com> wrote:
Michal,
VMs with attached devices are bound to host. When I use NFS I can migrate my VM and disks to another host.
Sure, it’s the usual speed/flexibility tradeoff with local nfs and migration you don’t get much performance Nothing is preventing you to set up the nfs server on one of the hosts and add it to oVirt, or perhaps gluster as Yaniv mentioned. For both speed and flexibility, we will one day have passthrough with migration support….just not yet:) Thanks, michal
On Wed, 2016-12-21 at 19:03 +0100, Michal Skrivanek wrote:
If disk performance is critical then there is an option to use direct access on local host using either PCI passthrough of a local storage controller or SCSI passthrough of LUNs.
participants (3)
-
Michal Skrivanek -
Pavel Gashev -
Yaniv Dary