I have three zpools on a server and want to add a mirrored pair of ssd''s for the ZIL. Can the same pair of SSDs be used for the ZIL of all three zpools or is it one ZIL SLOG device per zpool? -- This message posted from opensolaris.org
On 12/08/2010 07:27, Chris Twa wrote:> I have three zpools on a server and want to add a mirrored pair of ssd''s for the ZIL. Can the same pair of SSDs be used for the ZIL of all three zpools or is it one ZIL SLOG device per zpool?Only if you partition it up and give slices to the pools, however I personally don''t like giving parts of the same device to multiple pools if I can help it. The only vdev types that can be shared between pools are spares, all others need to be per pool or the physical devices partitioned up. -- Darren J Moffat
> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss- > bounces at opensolaris.org] On Behalf Of Chris Twa > > I have three zpools on a server and want to add a mirrored pair of > ssd''s for the ZIL. Can the same pair of SSDs be used for the ZIL of > all three zpools or is it one ZIL SLOG device per zpool?If you format, fdisk, and partition the disks, you can use the slices for slogs. (You can also implement it in some other ways.) However, the point of the slog device is *performance* so you''re defeating your purpose by doing this. People are always tempted to put more than one log onto a SSD because "Hey, the system could never use more than 8G, but I''ve got a 32G drive! What a waste of money!" Which has some truth in it. But the line of thought you should have is "Hey, the system will do its best to max out the 3Gbit/sec or 6Gbit/sec bus to the drive, so that disk is already fully utilized!"
> People are always tempted to put more than one log onto a SSD because > "Hey, > the system could never use more than 8G, but I''ve got a 32G drive! > What a > waste of money!" Which has some truth in it. But the line of thought > you > should have is "Hey, the system will do its best to max out the > 3Gbit/sec or > 6Gbit/sec bus to the drive, so that disk is already fully utilized!"That depends on your workload. I would guess most pools aren''t utilising their SLOGs 100%, since this will need pretty high sync-write usage, and typically, about 90% of the I/O on most servers is read. Vennlige hilsener / Best regards roy -- Roy Sigurd Karlsbakk (+47) 97542685 roy at karlsbakk.net http://blogg.karlsbakk.net/ -- I all pedagogikk er det essensielt at pensum presenteres intelligibelt. Det er et element?rt imperativ for alle pedagoger ? unng? eksessiv anvendelse av idiomer med fremmed opprinnelse. I de fleste tilfeller eksisterer adekvate og relevante synonymer p? norsk.
Thank you everyone for your answers. Cost is a factor, but the main obstacle is that the chassis will only support four SSDs (and that''s with using the spare 5.25 bay for a 4x2.5 hotswap bay). My plan now is to buy the ssd''s and do extensive testing. I want to focus my performance efforts on two zpools (7x146GB 15K U320 + 7x73GB 10k U320). I''d really like two ssd''s for L2ARC (one ssd per zpool) and then slice the other two ssd''s and then mirror the slices for SLOG (one mirrored slice per zpool). I''m worried that the ZILs won''t be significantly faster than writing to disk. But I guess that''s what testing is for. If the ZIL in this arrangement isn''t beneficial then I can have four disks for L2ARC instead of two (or my wife and I get ssd''s for our laptops). Thank you again everyone for your quick responses -- This message posted from opensolaris.org
> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss- > bounces at opensolaris.org] On Behalf Of Chris Twa > > My plan now is to buy the ssd''s and do extensive testing. I want to > focus my performance efforts on two zpools (7x146GB 15K U320 + 7x73GB > 10k U320). I''d really like two ssd''s for L2ARC (one ssd per zpool) and > then slice the other two ssd''s and then mirror the slices for SLOG (one > mirrored slice per zpool). I''m worried that the ZILs won''t be > significantly faster than writing to disk. But I guess that''s what > testing is for. If the ZIL in this arrangement isn''t beneficial then I > can have four disks for L2ARC instead of two (or my wife and I get > ssd''s for our laptops).Remember that ZIL is only for sync writes. So if you''re not doing sync writes, there is no benefit of a dedicated log device. Also, for a lot of purposes, disabling ZIL is actually viable. It''s zero cost which guarantees absolute optimal performance on spindle disks. Nothing is faster. To quantify the risk, here''s what you need to know: In the event of an ungraceful crash, up to 30sec of async writes are lost. Period. But as long as you have not disabled ZIL, then all the sync writes were not lost. If you have ZIL disabled, then sync=async. Up to 30sec of all writes are lost. Period. But there is no corruption or data written out-of-order. The end result is as-if you halted the server suddenly, flushed all the buffers to disk, and then powered off.
On Fri, August 13, 2010 07:52, Edward Ned Harvey wrote:> If you have ZIL disabled, then sync=async. Up to 30sec of all writes are > lost. Period. > > But there is no corruption or data written out-of-order. The end result > is as-if you halted the server suddenly, flushed all the buffers to disk, > and then powered off.With the proviso that you should ideally be using a version of OpenSolaris later than snv_128 which allows you to go back a previous uberblock/txg in case the more recent one(s) are not viable: "zpool recovery support" http://arc.opensolaris.org/caselog/PSARC/2009/479/ "need a way to rollback to an uberblock from a previous txg" http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6667683 http://www.c0t0d0s0.org/archives/6067-PSARC-2009479-zpool-recovery-a.html If you''re at ZFSv20 or later, you''re pretty much guaranteed to have this functionality: http://hub.opensolaris.org/bin/view/Community+Group+zfs/20 I''m hoping this, and slogs removal are incorporated into Solaris-proper soon: http://hub.opensolaris.org/bin/view/Community+Group+zfs/19