| Author |
Message
|
| CuriCAT |
 Posted: Mon Aug 10, 2009 8:30 pm Post subject: Performance tuning at EG level & at BRK level |
 |
|
Voyager
Joined: 26 Sep 2006
Posts: 82
|
We have a message flow with following design
MQ Queue --> Mapping ----> MQSIFileOutput
We did few performance tests.
Environment we tested : pSeries AIX, 2 CPU - Flow processed one million records in 40 mins.
We run the test for one million records in a much powerful system: 7 CPU, pSeries AIX. there is no significant performance improvement, It processed 1 million record in 35 mins. We were surprised to see the same performance in a better configuration.
Do you guys have any suggestion? In EG level or at Broker level the environment can be tuned to gain better performance?
http://publib.boulder.ibm.com/infocenter/wmbhelp/v6r0m0/index.jsp?topic=/com.ibm.etools.msgbroker.helphome.doc/help_home_msgbroker.htm
ac55120_ |
|
| Back to top |
|
 |
| elvis_gn |
| Posted: Mon Aug 10, 2009 9:28 pm Post subject: |
|
|
Padawan
Joined: 08 Oct 2004
Posts: 1905
Location: Dubai
|
Hi CuriCAT,
Improving hardware cannot simply keep improving the performance of any application...every piece of code needs some amount of processing time, just by giving it more CPU does not mean it would thread the operation to keep improving performance, UNLESS you have a multi-threaded CPU sensitive code.
If you want to bring the flow to the optimum performance level you need to look at your code...other than that you could look at adding additional instances.
Btw, 1 million in 40 min...what more are you expecting 
Regards. |
|
| Back to top |
|
|
| zpat |
| Posted: Mon Aug 10, 2009 9:55 pm Post subject: |
|
|
Jedi Council
Joined: 19 May 2001
Posts: 5866
Location: UK
|
A single flow instance can only use one CPU in any case.
To improve throughput, use multiple execution groups with the same flow (assuming it can run like that).
For heavy I/O look at faster disks (higher RPM) with large write caches, or possibly the network if the disk is not locally attached.
Look at the IBM performance support pacs. |
|
| Back to top |
|
|
| fjb_saper |
| Posted: Mon Aug 10, 2009 11:29 pm Post subject: |
|
|
Grand High Poobah
Joined: 18 Nov 2003
Posts: 20756
Location: LI,NY
|
If you need more throughput in the same configuration,
besides looking at scaling instances in the e.g. and running the flow across multiple e.g.s have you thought about checking the difference between a mapping node and an esql node doing the mapping with the use of references ?
It appears this difference is much less in 6.1. We found it significant enough in 6.0 to use ESQL nodes for our mapping.
_________________
MQ & Broker admin |
|
| Back to top |
|
|
| rbicheno |
| Posted: Tue Aug 11, 2009 3:59 am Post subject: Performance tuning at EG level & at BRK level |
|
|
Apprentice
Joined: 07 Jul 2009
Posts: 43
|
As suggested sounds like you hitting a bottleneck that you donot understand, you should monitor resources like i/o and cpu as well as throughput to get a better understanding. From your description i suspect your bottleneck is either that you donot have enough additonal instances set or that you are saturating the disk on the fileoutput node (this would depend on the speed of disks and your message sizes).
Also seeing as you are using a mapping node and want to improve performance i would suggest you look at this supportpac which i have used sucessfully several times to good affect:
http://www-01.ibm.com/support/docview.wss?rs=171&uid=swg24017156&loc=en_US&cs=utf-8&lang=en |
|
| Back to top |
|
|
|
|
