Optimizing performance with SLOG/L2ARC

[PK1048]

On Aug 19, 2015, at 12:02, Schweiss, Chip <chip at innovates.com> wrote:

> On Wed, Aug 19, 2015 at 10:29 AM, PK1048 <paul at pk1048.com> wrote:
> 
>> 
>> Please note that, depending on your workload, an SSD may _not_ be any
>> faster than a HDD. I am in the process of rebuilding a file server that
>> exhibited poor NFS SYNC write performance. Yet it had a mirrored pair of
>> SSDs. Unfortunately, those SSDs had _worse_ write performance than an HDD
>> for small (4 KB) writes. Based on recommendations from the OpenZFS list I
>> have a pair of Intel 3710 SSDs coming in to try, they are supposed to have
>> much better write performance (at all block sizes) and much better
>> reliability long term (10x full disk writes per day for 5 years). I’ll know
>> more once they arrive and I can test with them.
>> 
> 
> Pure SSD pools still need a log device.

Sorry I was unclear, I was NOT suggesting a pure SSD pool.

>   ZFS doesn't play well with the
> ZIL on the pool with SSDs.   Even an SSD of the same type as the pool
> devices as the log device will fix the latency problem and throughput
> problems.

If your load is sync writes then you decidedly want a LOG device, even if it is the same type as the devices in the pool. For the reasons others have posted.

> It seems counter-intuitive but a very real problem, there is a long thread
> about this on the Illumos ZFS list.  If you don't believe it, turn off sync
> on your SSD pool and performance will skyrocket.

But remember to turn it back on after you test so that you don’t break posix sync behavior and raise the possibility of loosing writes in flight. I am horrified at the number of posts on the Internet that tell you to simply disable sync to fix sync performance issues (VM images accessed via NFS being a very common one).