Tracking down high COMM_WAIT times

[rowl]

In some testing I am doing I am experiencing unusual COMM_WAIT times. I am curious what one looks for to track down the cause of this. In this test I have an AIX DB2 server that is also running TSM 5.3, and backing up directly to a VTL.

During the restore I see the jobs are spending 75-80% of their time in a COMM_WAIT state. So the 2 hour restore is waiting for 90 minutes.

What would contribute to such high COMM_WAIT times?

Thanks,
-Rowl

 

[SANdmann]

Godd Question - Still no Answer

Hi!

I am trying to do restore tests and i get the same "problem".

same of the restores have high mediawait - why not.. no drive, volume not available
but some also have VERY high COMM_WAIT >5000 up to 9000... That would be 150 minutes???? Is that just normal for File Nodes with large amout of small files?

Any Help?
Any Ideas?

Thanks

 

[moon-buddy]

Hi!

I am trying to do restore tests and i get the same "problem".

same of the restores have high mediawait - why not.. no drive, volume not available
but some also have VERY high COMM_WAIT >5000 up to 9000... That would be 150 minutes???? Is that just normal for File Nodes with large amout of small files?

Any Help?
Any Ideas?

Thanks

Maybe normal if you have very limited tape drives and a single restore instance.

Should you have the luxury of getting more tape drives, increase the restore instances and you will have faster restore times.

 

[SANdmann]

Comm_wait

Yes,

but if the MEDIAWAIT ist low, and the COMM_WAIT is high...
Then we have a problem regarding network capacity?
Or what is meant with communication layer?
If its the client... nothing i can do is dont administrate it... :/

Found the following:
Communications Wait (CommW, commwait) "Sess State" value in 'Query SEssion'
for when the server was waiting to
receive expected data from the client or
waiting for the communication layer to
accept data to be sent to the client.
An excessive value indicates a problem
in the communication layer or in the
client.
Recorded in the 23rd field of the
accounting record, and the
"Pct. Comm. Wait Last Session" field of
the 'Query Node Format=Detailed' server
command.
See also: Idle Wait; Media Wait; RecvW;
Run; SendW; Sess State; Start

 

[n9hmg]

You know, this sort of sounds like a database performance issue - taking a long time to build the restore session. I think that gets reported under "CommW".

 

[rowl]

I have still never found any good information on this topic, but I have another issue where this looks like a problem. I am looking at a Linux client backup that over a 30 day period runs an average of 19.4 hours with an average COMM_WAIT time of 5.6 hours. The backup is around 1TB, 200,000 items examined and around 4000 backed up.