[zfs-discuss] should SSD zil improve sequential write speed

Ryan How rhow at exemail.com.au
Sun Jun 24 22:24:37 EDT 2012


You could try adding a sync flag to dd to just test it out.

dd bs=128k count=10000 if=/dev/zero of=test conv=fdatasync
dd bs=128k count=10000 if=/dev/zero of=test oflag=dsync

fdatasync should normal write and sync at the end
dsync should sync every block


On 25/06/2012 7:57 AM, Ty Newton wrote:
> Thanks for that info.  I didn't realize it only impacted specific programs.
>
> I'm guessing dd doesn't do this and Samba probably doesn't either.
>
> Is there a program to test zfs performance?
>
> Ty
>
> "The content of your character is your choice.  Day by day, what you do is what you become. Your integrity is your destiny"
>
> On 25/06/2012, at 9:08 AM, Matthew Robbetts<wingfeathera at gmail.com>  wrote:
>
>> Hi Ty,
>>
>> Someone else on the list can probably give you a more detailed answer, but the short/rough answer: the ZIL will not improve general write performance - that is not what it is for. The ZIL serves as a cache only for *synchronous* writes, which are a special write performed only by applications needing a guarantee that the write has not only been accepted by the storage subsystem, but is actually *on persistent storage*. The application will block until the write is complete, making it entirely dependent on the speed of the storage. The major example people mention is databases, which need this guarantee so that their data don't get corrupted by a crash or power outage that occurs while a clever disk system is reordering its queued writes (for performance). A ZIL allows ZFS to guarantee the application that the write has genuinely been performed, but also to service the write much more quickly than if only spinning disks were present.
>>
>> I believe that synchronous writes are buffered immediately into the ZIL and are then farmed off to spinning disks at a convenient later time.
>>
>>
>>
>> On 24 Jun 2012, at 23:48, Ty Newton wrote:
>>
>>> Hi,
>>> I've just put a intel 320 SSD into my zpool hoping to improve write performance but it doesn't seem to have made any difference.
>>>
>>> I have a raidz2 zpool on SATA 2, 1TB green WD drives (9 of them).  I can get about 100MB sustained write performance.  After I put the ZIL in I get about the same.
>>>
>>> I'm running Ubuntu 12.04; 8GB ram, quad core core2 CPU.  I am testing with a dd command to create a zero filled 100GB file on the array and looking at zpool iostat.  All the drives give the same performance numbers.
>>>
>>> Strangely the zil doesn't tend to report anything but 0 in the iostat output.  Is there a way to be certain that it is getting used.
>>>
>>> Any help is greatly appreciated,
>>> Ty
>>>
>>> "The content of your character is your choice.  Day by day, what you do is what you become. Your integrity is your destiny"





More information about the zfs-discuss mailing list