zondag@xs4all.nl (Jan-Evert Zondag) wrote:
Some additional things you might look at:
Does your version of UNIX support kernel asynchronous IO? If so, then
informix allocates one virtual processor per CPU to handle IO. Here
again, the more cpus you have the better your performance will be.
Also, are you using any kind of RAID setup for your rawdisk? This can
have a tremenedous negative impact on throughput, especially for
writes. We use a RAID 5 setup. Writes are relatively slow but our
reads (98% of what we do, scream). Monitor your SAR output during your
testing and examine you onstat -p output to try and identify your
bottle necks. Use onstat -D to monitor actual chunk IO. Are you using
data partitioning? If you are not using a RAID configuration, then you
might want to consider it. If you are using RAID, then avoid
partitioning.
>In article <3qg2a5$1d1e@firehose.mindspring.com>,
> ekelley@mindspring.com (Ed Kelley) wrote:
>>One thing you might look at is the PDQPRIORITY and MAXPDQPRIORITY
>>paramteres in your config file. If both are set to 100, then a query
>>will use 100% of the system's resources to complete that query. A
>>better alternative is to set PDQPRIORITY to "off". Then control the
>>actual usage on a job by job bases using the environmental variable
>>PDQPRIORITY.
>>
>>Before I did this, I had the same problem with 7.1 on an AT&T 3555
>>server. Now, it smokes!
>>
>>Something else to think about is 7.1 is heavily multithreaded. A dual
>>processor system does not really take advantage of this. More cpus
>>greatly improve performance.
>>
>thanks for your fast response. We will run a few extra tests. We had
>PDQPRIORITY = OFF but we did not know it could be controlled by setting an
>environment variable.
>But... our performance degradation also occurs when a number of update
>processes is running.
>---------------------------------------
>Met vriendelijke groet (dutch for: with kind regards)
>Jan-Evert Zondag
>email: zondag@xs4all.nl
Ed Kelley, Senior Specialist
AT&T Network Systems - CIO
Email: ekelley@mindspring.com or ekelley@attmail.att.com or ekelley@msn.com or ekelley@ichange.com
WEB page: http://www.mindspring.com/~ekelley/html/ekelley.htm