General
cancel
Showing results for 
Search instead for 
Did you mean: 

measure network performance and latency Oracle RAC

Mohd Azahari
Advisor

measure network performance and latency Oracle RAC

Hi there,

does anybody knows how to measure network latency/performance in Oracle RAC environment between nodes/clients.

I noticed there are ORA-3136 in my alert.log which indicates network issue but I do not how to proof it.

any help appreciated.
5 REPLIES
Steven E. Protter
Exalted Contributor

Re: measure network performance and latency Oracle RAC

Shalom,

Network issues are network issues.
http://www.hpux.ws/?p=6

There is an HP-UX performance gathering script.

If there is a network issue, you will spot it.

The tool runs background and does not use much resources.

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
Mohd Azahari
Advisor

Re: measure network performance and latency Oracle RAC

hi,

the link doesn't work
Viktor Balogh
Honored Contributor

Re: measure network performance and latency Oracle RAC

though, the link works. you should click it instead of copy-paste it.
****
Unix operates with beer.
Mohd Azahari
Advisor

Re: measure network performance and latency Oracle RAC

thanx all for helping
Hein van den Heuvel
Honored Contributor

Re: measure network performance and latency Oracle RAC

I suspect that this is goign to be a DBA challenge first and foremost, NOT an SA task.

Did you start with a Google for "ora-3136" ?
You'll quickly stumble into lots of topics on the matter such as:

http://forums.oracle.com/forums/thread.jspa?messageID=1612957&tstart=0
Which amongst other reads:
"WHEN this error occurs in a DATAGUARD environment, normally you don't have to worry about it. It happens, when connection between primary and standby side is lost temporarily or not yet established. Of course it should not happen too often."

I suspect this is not a performance issue per-se but more and up/down, process there/process gone kind of situation. Not measured in milliseconds or packets per second but in on/off events.

And even if further investigation starts to point to raw performance after all then with all due respect to SEP, those system counters aren't going to tell you much to begin with.
Sure they are useful to get an impression of what the system is dealing with but if you really want to know what is troubling Oracle, then you should ask Oracle what seems to be the problem.

You HAVE to approach this from the Oracle angle first and seek confirmation from system stats as seems useful.

Check out ADDM, AWR, STATSPACK.
Look for RAC counters like ""GCS CR BLOCKS RECEIVED" and "GCS CR BLOCK RECEIVE TIME" and many more in GV$SYSSTATS.

Read documents like (just one of many)
http://www.oracle.com/education/america_eblasts/fy09/q1/rac_tipsandbestpractices.pdf

good luck.

Hein van den Heuvel
HvdH Performance Consulting