[OmniOS-discuss] Resilver speed and record size?

[Dan Swartzendruber]

I have a pool with 3x2 mirrors.  SAS nearline drives.  One of them died,
so I replaced it, and it resilvered overnight.  Another drive is showing
40 grown defects, so I am suspicious.  I want to replace it and reformat
it to see if any problems crop up.  So.  I plugged in a new drive and did
'zpool replace tank foo bar', and it's going along just fine.  Kinda slow
though, compared to scrub.  It started at about 5MB/sec and is now up to
13MB/sec, which is pretty slow.  I've been googling and find posts various
places that the resilver speed depends to some extent on record size
(apologies if I am mis-stating this.)  A good chunk of the data (180GB) is
in a zvol serving up storage to an esxi server for virtual machine disks
(zvol, so defaulting to 8K).  Probably twice this is in nfs-exported
shares for veeam backup (defaulting to 128K).  Originally everything was
nfs-shared, and I seem to recall resilvering being faster.  Am I out to
lunch here?  Any thoughts or suggestions as to what I should do ('nothing'
is a valid reply, of course) would be much appreciated.