If you are unable to create a new account, please email support@bspsoftware.com

 

News:

MetaManager - Administrative Tools for IBM Cognos
Pricing starting at $2,100
Download Now    Learn More

Main Menu

Server setting optimization

Started by i-guy, 02 Nov 2005 03:59:07 PM

Previous topic - Next topic

i-guy

Hi.
Curious what people have seen regarding optimizing windows setting for quicker response (esp w/publishes).
Esp curious about...
virtual memory settings (By default, the virtual memory paging file  that is created during installation is 1.5 times the RAM.) ,
Memory Usage (Programs vs  System Cache),
and Processor scheduling (Programs  vs Background services).

Cognos only wants to give me the typical minimum system requirement spiel.

thx.

andrewbho

#1
There are several things you want to make sure.  I am not sure who performed your base installation/configuration but you want to make sure that hyperthreading is turned off, you want to make sure that you have at 2 gbs of RAM per CPU, most importantly make sure that all of your servers are within the same subnet.  You can adjust server polling and ping times, which will reduce your network traffic.  If all else fails, purchase additional job servers to improve performance linearly.  When you run a  publish a .out file is created and pushed to the DB server.  The virtual memory settings will not do too much since publishes are more CPU intensive than memory intensive; however, most of the virtual memory that I've seen is about double the ram on the server(s).  I would always set it to let it choose automatically and whatever the default is.  If you run publishes from Analyst, you will only use 1 CPU since Analyst is a single threaded application.

Cognos will always give the min. amount of resources.  It helps them sell software because it reduces the total cost of ownership.

Hope this helps.  Let me know if you have any other performance related questions.

[email address removed - sorry Andrew, for security/spam purposes only]

admin1234

Some other things you may want to consider:

* CPU is ultimately is where you'll gain most because the job server has to retrieve the XML GUIDs from the datastore, parse them into human-readable data and then push it back to the publish container.  Get the fastest CPU on the market.  Do not use Hyperthreading.

* Start to consider adding more job servers to the job server application cluster.  The more job servers you have the faster the processing.  Performance is almost linear.

* Use RAID0 or RAW disk for your TEMP volume and direct %TEMP% and %TMP% to here.  Of course, different physical disk.  Multi I/O controller is best.

* Publish to a separate container.

* Exclude the %temp% directory in your viruscan if you have this turned on.  The publish process retrieves each node item into this area and then pushes it back to the publish container via the bulk load program (bcp for SQL2k, sqlldr for Oracle).  Viruscan programs can slow this down.

HTH,
admin1234

ykud

QuoteDo not use Hyperthreading
admin1234, why that? How HT diminishes performance?

andrewbho

HT dimenishes performance on apps like Planning bc Planning is made to be "efficient" and when it runs job it maxes out the CPU to 100% utilization.  When HT is turned on you essentially have the processors running at 200% + normal overhead, which will cause your jobs to process slower and fail.  Make sure that you always turn it off.

chriso

Hello CenterStone,

Switching the HT off, does it apply to client PC as well? or the job server only?

cheers,
/chriso

andrewbho

Job servers only.  Most PC's do not have a hyperthreading option available.  You turn it off via the server bios.

Hope this helps.

chriso