ZVOL recommended block size

[newlink]

Is the ZVOL 8 kB block size the most performant option (iSCSI+ZFS) with a VMFS-5 volume?

VMFS-5 datastores use a 1MB block size and 8kB sized sub-blocks, so the default 8 kB block looks like a pretty reasonable and logical option to increase IOPS (while probably sacrificing a little bit the total data burst) but I haven't seen accurate comparative tests so far with bigger blocks (64 kB, for example) and Freenas

 

[cyberjock]

I leave the default and use that. In the big scheme of things, from personal experience, having hardware that can do the job is more important than setting a block size. Usually this means lots of RAM and an L2ARC.

Of course, if you do something silly like do 512byte blocks you can expect performance to take a serious nosedive.

 

[zambanini]

it is more important to change the iscs parameters. search within the forum, somebody did some benchmarks with vsphere and mpio

edit:
https://forums.freenas.org/index.ph...hen-single-path-for-writes.22475/#post-150037
i understand, that you already posted there.

 

[newlink]

Too big blocks are out of question, but a lot of commercial SAN use a default 64 or 128 block size, probably because the old VMFS-3 standard used a 64 kB block. How about partition aligment between different layers (LUNS -> VMware -> VM disks)? Is it important?

 

[newlink]

it is more important to change the iscs parameters. search within the forum, somebody did some benchmarks with vsphere and mpio

edit:
https://forums.freenas.org/index.ph...hen-single-path-for-writes.22475/#post-150037
i understand, that you already posted there.

In my experience IOPS limit = 1 makes the real difference, that was a really good and documented post