From: Joe Fletcher (fletch@csl.psychol.cam.ac.uk)
Date: Sat Apr 13 2002 - 06:50:51 EDT
Thanks go to Alan and Joerg for responses with the beer & pizza going to
JC Faul for suggesting some tuning changes. I dropped ubc_maxpercent from
100 to 50 and got an immediate improvement with I/O rates going up to
around 10Mb/s.
Curiously though the DS20 still has ubc_maxpercent set to 100.
Some more digging to be done but I'm happy with the gains.
Cheers
Joe
---------- Forwarded message ----------
Date: Fri, 12 Apr 2002 16:13:14 +0100 (BST)
From: Joe Fletcher <fletch@csl.psychol.cam.ac.uk>
To: tru64-unix-managers@ornl.gov
Subject: Slow I/O?
Hi,
I asked about this before but it's come back to bother me again.
Scenario:
ES40, 4x500, 4GB, 2x KZPBA-CB U/W/D controllers talking to
RA3000s (HSZ22). The RAs have RAID5 sets across 6x36GB drives. ADVFS
Filesets mounts on users2 and users3 say.
DS20e, 2x833, 2GB, KZPBA-CA U/W with a couple of IBM 36gb drives in a box.
1 ADVFS volume across the two disks. Fileset mounted on users1 say.
If I run "vdump -0 -f /dev/null /users3" on the ES40 iostat shows transfer
rates of about 1-1.5Mb/s.
If I run "vdump -0 -f /dev/null /users1" on the DS20e iostat shows transfer
rates of about 15-20Mb/s.
Can anyone tell me what I'm missing on the ES40 that's making it so slow
relative to the DS20? System tuning is ok as far as I can see. sys_check
doesn't suggest anything out of place. The book on the RA3000 indicates it
should be able to sustain 25Mb/s reads on a RAID5 set.
Both systems are running 5.1 patched up to current revs. I don't believe
the difference can be attributed to the EV67s in the DS20e compared to the
EV6s in the ES40. For info users3 on the ES40 is on a newly created file
domain. Fragmentation is minimal. I get the same results on users2 on the
ES40 which is an older domain created originally under v4.0f.
Any ideas anyone?
TIA
Joe
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:48:38 EDT