Page 1 of 1

Poor performing DataStage server - 'hanging' connections

Posted: Mon Jun 01, 2009 7:12 am
by sjordery
Hi All,

We are running an instance of DataStage 7.5 on a Linux box with 4 CPUs.

The box is accessed both locally in the UK, and by a team in India. Since the team in India came on line, the performance of the box seems to have degraded badly. The slow down happens over a couple of days, and can normally be resolved by stopping/restarting DataStage - and clearing 'hanging' processes.

I have had the Unix admin monitor the box during these slow times and all appeared to be fine - CPU not being hammered, plenty of disk space free etc.

However, what we do have is several (sometimes tens of) sessions appearing as ESTABLISHED (checked using netstat -na | grep 31538) - even after everyone has logged off.

Is it possible that it is these sessions, or something associated with them, that causes the loss of performance?

I am told that the developers in India frequently get 'broken connection' messages requiring them to log back in - so this is likely the source of these extraneous sessions.

Thanks in advance.
S

Posted: Mon Jun 01, 2009 8:35 am
by ArndW
Do you have a timeout set from your administrator? Do the remote developers login directly or do they go through an application such as Citrix?

Posted: Mon Jun 01, 2009 9:04 am
by sjordery
Thanks for your reply.

They log into directly to the server. They have also complained that at times it is taking 5 mins+ to refresh logs in Director, so the network speed transferring data from client to server might be slow.

Inactivity timeout is set to 'Do not timeout'

Thanks,
S

Posted: Mon Jun 01, 2009 9:06 am
by chulett
Them having their own local copies of DataStage is a bad idea, you really need to look into setting up a shared Citrix-type server locally and have them connect thru that. Would go a long ways towards solving the speed issue on their end and the disconnects.

Posted: Mon Jun 01, 2009 9:10 am
by sjordery
Thanks Craig. I will definately push those suggestions forward.

I know that it is impossible to make accurate diagnosis without looking at the server yourself, but do you think that the general slowing down of the server could be attributable to these disconnects?

Thanks again.

Posted: Mon Jun 01, 2009 9:13 am
by chulett
More than likely, yes. Do you have the 'deadlock daemon' enabled on the server? If not, you should, it would help as well as one of the functions it performs is to clean up disconnected sessions like that.

Posted: Mon Jun 01, 2009 9:21 am
by sjordery
Thanks Craig.

I'm not aware of a deadlock daemon, and so will have a chat with the unix admin to see if we can get anything put in place.

Thanks again for your time, much appreciated.

Posted: Mon Jun 01, 2009 9:38 am
by chulett
FYI, it's a DataStage 'tool' not anything generic UNIX. Many discussions here, for example this one recently.

Posted: Mon Jun 01, 2009 9:58 am
by sjordery
Perfect, thanks.