cancel
Showing results for 
Search instead for 
Did you mean: 

regarding true cluster

N/A

regarding true cluster

Is there any document describing how to tune the system for better performance

one of orur customer shifted from normal tru64 to unix tru64 cluster and the performance of the product is 4 times slower than the former

for more information customer is using an edi product of hp named dec/edi
6 REPLIES
Rob Leadbeater
Honored Contributor

Re: regarding true cluster

Hi,

It's important to realiase that a clustered environment isn't necessarily going to give you any improvements in performance. Particularly if the applications don't understand how clusters work.

I've no experience of dec/edi but looking at the website http://www.decedi.com it would appear to be quite an old product. I would ask the software vendor how best to tune the OS for their software. As that vendor is HP, you should hopefully get a sensible answer !

Cheers,

Rob
N/A

Re: regarding true cluster

BUT THE PRODUCT IS CLUSTER COMPATIBLE
cluster routines are there for supporting the cluster environment IN DEC/EDI
Rob Leadbeater
Honored Contributor

Re: regarding true cluster

Just because it's compatible doesn't mean it's going to run faster...

If the product runs in failover mode, ie it normally just runs on node A in the cluster and restarts on node B in case of a failure of node A, then you probably will see worse performance. You will of course have better availability.

Cheers,

Rob
Hein van den Heuvel
Honored Contributor

Re: regarding true cluster

Why did they go to a cluster?
What is the expectation for the cluster? Increased availabilty or increase performance?
What OS and Application versions are involved?

What is involved the test? SMTP? FTP? Mime? A database connection?

The attachment shows a 3 month old performance graph with no indication of what might be happening other then the name of a product.

It also shows 40% user mode and 60% system time. So go figure out what the CPUs are doing while system mode and you'll be on you way to an explanation.

Is this the same problem as you questioned about a month ago?
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=1134220

What progress was made?
Where are the basic trouble shooting skills? Where is the common sense? Is this for real, or just farting around? If this is a serious problem with real money involved and the reputation of real product at stake, then please get serious help!

If I had to make a guess, then I'd still suspect that this application is really only active on a single node and it is writing data to files (audit logs?) on a mountpoint which is 'owned' by the other node. Has that been checked in the last month?

- use cfsmgr to visualize the mountpoint ownership and then to force the ownership to the node running the test.

- If that solves the problem then adapt the CAA / failower scripts to move mountpoint ownership.

- If that was nto the problem then go analyze the system time usage: dcpi, or good old kprof.

Not having already provided that info here is embarrasing imho.

Good luck!

Hope this helps some,
Hein van den Heuvel (at gmail dot com)
HvdH Performance Consulting

Ivan Ferreira
Honored Contributor

Re: regarding true cluster

You should check the ownership of the file systems as menthioned above with cfsmgr. You can also check statistics with this command and relocate the file systems based on access patterns.

Also, what is your cluster interconnect technology? Memory Channel or LAN Interconnect?

I had several performance problems with LAN Interconnect, mostly because of autonegotiation settings.

Por que hacerlo dificil si es posible hacerlo facil? - Why do it the hard way, when you can do it the easy way?
N/A

Re: regarding true cluster

regards,
ajaydec