1833860 Members
2222 Online
110063 Solutions
New Discussion

sar/vmstat vs glance

 
SOLVED
Go to solution
Regina Mitchell
Frequent Advisor

sar/vmstat vs glance

I am working with Oracle DBA to analyze performance problem. I have a Superdome running 11.i OS. the question I have is elow.

The SAR & VMSTAT are showing Wait IO upto 28% when I run these utilities on server. However, glance and top are not showing io waits that high. Idle time in SAR averages around 40% whereas in TOP it is around 60 to 65%. I would like confirm which source is correct.

Thanks

All replies are appreciated.
5 REPLIES 5
Steven E. Protter
Exalted Contributor
Solution

Re: sar/vmstat vs glance

Shalom,

Sorry, you probably can't confirm which source is correct.

sar and glance collect their data in different ways and at different intervals. Sometimes even the calculation methodology is a little different.

Therefore coming to a conclusion about which data is correct is probably impossible, since more than likely both data sets are correct based on the criteria I outlined above.

I don't have glance in all locations and try and use a standardized set of sar tools for this information, and they've never pointed me wrong as far as finding performance problems.

http://www.hpux.ws/system.perf.sh

Adjustable, runs background and thats production code I just posted up there, tried and tested world-wide.

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
Zeev Schultz
Honored Contributor

Re: sar/vmstat vs glance

I'd vote for Glance as it comes to accuracy. It uses midaemon collected set of data,as oposed to
ps,top,sar,vmstat which use kernel counters (pstat). And obviously would suggest OVPA (former measureware) as a way to collect data and analyze later.
So computers don't think yet. At least not chess computers. - Seymour Cray
Bill Hassell
Honored Contributor

Re: sar/vmstat vs glance

sar and vmstat report very crude measurements and technically, an single I/O wait percentage is almost meaningless. You want to look at specific disks and look for large queue lengths (more than 2 or 3), as well as avwait and avserv times that are very high (double digits or more). If the numbers are all low during a busy period, the problem is probably Oracle not using enough SGA, unbalanced indexes, full or partial searches in the SQL code, etc.


Bill Hassell, sysadmin
Regina Mitchell
Frequent Advisor

Re: sar/vmstat vs glance

All,
Thank You for your replies. The explanation did help understand the discrepencies.
Regina Mitchell
Frequent Advisor

Re: sar/vmstat vs glance

The answers provided a better understanding to my question.