ESXi 5.5 ISCSI FreeNAS 9.2.0 slow random write speed (read is ok)

[Thomymaster]

Hi

OK but this still doesn't explain the behaviour i discovered above.

I read your conclusion but:

5) Sufficient ARC or L2ARC to hold the entire working set (blocks read in a given period of time, like 1 hour or 1 day),
6) Something for SLOG.

I don't use a L2ARC because it gives me now performance benfit neither SLOG as it should not be needed when running ISCSI (this is what i've been told).

 

[jgreco]

You should be running sync=always for iSCSI on ZFS for VM storage, at least if you care about the contents of your VM disks if something goes awry on the filer (power loss, crash, etc). In that case, SLOG becomes rather important.

The last several months have seen me too busy to play much, so you get the commentary I have rather than the solutions you wish I had, sorry. I'm very interested in the problems associated with VM storage but it is complicated to fully understand all the interactions.

 

[ZFS Noob]

Is there any explanation for this?​

I may be misunderstanding you, but it sounds like you moved data from one volume to another, which will have defragmented it in the process. That could be your explanation. Move everything off the datastore and move it back, then see if your performance improved.

 

[Thomymaster]

OK i see i have to buy one (or 2 for mirrored SLOG) SSDs, but why does the vDev behave better than the file-based approach (ZFS dataset with one file for the istgt)?