zfs discuss - Sep 2009 - I/O performance problems on Dell 1850 w/ PERC 4e/Si RAID1 (Sol 10U6)

I''ve got Solaris 10 U6 with a ZFS boot/root installed on a Dell 1850
here, patched up as of this morning (kernel 141415-10).  The system is running
with hardware-mirrored disks, via the LSI MegaRAID controller, a PERC 4e/Si.  4G
of RAM, one dual-core 3.4GHz CPU.

The system is a simple mail hub, and it does the task very well.  Most files
exist only very briefly, and thus rarely touch the real disk.  However, whenever
it DOES access the disk, I see long wait times, sometimes four to ten second
hangs.  For example, ''df -h'' will print a single line and hang
for about five seconds, and then complete.  Subsequent runs complete near
instantly.

There will normally be a pause whenever running a "new" command or
opening a new file.  Running it again is fine.

Scp''ing a large file to the system results in 50 to 60 megs worth of
transfer, and then a stall for five to ten seconds, then continuing.

I''ve seen others complain about ZFS performance on the MegaRAID
controllers, and I''ve tried some of the suggestions in those threads...
dialing down zfs_vdev_max_pending (currently 35, I''ve tried 15 and even
4) and turning off prefetch, all via mdb -kw... it doesn''t seem to make
a difference.

Any suggestions as to where I can try to figure out what''s going on
here?
-- 
This message posted from opensolaris.org