From: Rob De Langhe (Rob.DeLanghe@telindus.be)
Date: Fri Aug 20 2004 - 06:07:46 EDT
all,
we have a bizarre performance problem here with the I/Os sent to an
EMC-Clariion CX600 array from a SUN-V480 over FC-AL (QLogic HBAs) :
- the machine is an Oracle (9.2.0.2) server, 2x900MHz, 6GB RAM
- the output of "sar -u" (long intervals of 15 mins) learns us that the
machine is average 60-80% busy waiting for I/O completions, still > 0% but <
5% idling
- the output of "sar -d" (long intervals of 15 mins) learns us that 2 LUNs
(each RAID-5, 935GB) are mostly read from, occasionally written to, each LUN
bares a separate database instance, with average wait-times of 0 ms, avg
service-times of < 5 ms
- comparing this with the output of "iostat -x" (short intervals of 5 secs)
for the same 2 LUNs, where kr/s (kilobytes read) has peaks of 50 MBps and
service-time of 20ms and %b ("pct of time the disk is busy") = 100%
=> found here and there articles on the web where the max throughput per LUN
per Storage Processor (=array-controller) is in this order of 50-60 MBps, so
that would be normal behaviour
BUT: most of the time the activity to the LUNs is in the order of 500-1000's
kr/s, with service-times of 5-10 ms, but still the %b is 70-100% !!!
And indeed this reflects in the high Waiting-for-IO percentage of the machine,
and slow performance of the database instances.
EMC claims this would be normal behaviour: I accept this for the peaks of 50
MBps, but surely not for the 500-1000 KBps reads !! Come on, a dumb tape
achieves higher throughput than this !
Any ideas ??
TIA
Rob
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:29:19 EDT