Folks,
I just joined this forum and this is my first ever post. I am doing a
performance analysis of our mission critical application. Below is the salient
configuration information:
- OS is Solaris9
- Server is a SunFire 20k domain with 20 dual-core SPARCIV+ processors
- Storage is EMC's DMX-3000
- The server has four HBAs using DMP for multi-pathing
- Veritas VM 4.1
- This is an Oracle 11.5.9/9.2.0.6 (64-bit) application and the database size
is 1.5 TB. The databases is using Oracle's ODM (Oracle disk manager) interface
for I/Os. The application is hybrid in nature with mostly batch and does a lot
of I/Os
As part of statistics gathering, I collected stats from iostat, vmstat and sar
at two-seconds interval. The analysis of the asvc_t and wsvc_t statistics from
iostat shows that approximately over 95% of the I/O had the average response
time (asvc_t+wsvc_t) of around 10 ms or less. This looks quite good to me.
However, at the same time, vmstat's "b" column consistently shows numbers that
average around 12 but fluctuate into 30s sometime. This is where I am having a
hard time understanding it, that is, iostat looks good but vmstat is reporting
blocked I/O threads.
Can someone please explain what should be primarily looked at when analyzing
I/Os on a system; I am thinking that iostat is the tool to consult. But then
what is the significance of the "b" column in vmstat and how much should one
worry about this value.
I would like to point here that the "r" column from vmstat and "runq-sz" from
sar mostly have "0" values and the CPU usage (usr+sys) is also not too high.
Any help will be appreciated
Thanks
This message posted from opensolaris.org
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org
