Showing results for 
Search instead for 
Did you mean: 

Informix on HP

Praveen Bezawada
Respected Contributor

Informix on HP

We are using informix for one of our application running java1.3 on HPUX11 ( L2000).
We find that database is bottlenecks as observed from variuos java profiling tools.
Also observation from glance shows RSS value of the informix process is quite low.
We suspect that low memory utilisation by informix problem could be causing the problem.
What are the configuration that influence informix memory utilisation.
We are using raw disk for informix.
Thanks in advance
Krishna Prasad
Trusted Contributor

Re: Informix on HP

Well not being an Informix expert but I do use other relation databases. I would ask you how much memory is Informix asking for?

Other databases use a config file that allocates a shared memory segement, and or a data dictonary in a shared memory segement.

The OS needs to have the following kernel settings configured to be large enough for what ever the database is asking for.

Also, the correct amount of swapspace needs to be configured. This is usally 2 * the amount of memory you have in the box. Although some applications may recommend 3 * the amount of ram.

I hope this is of some help
Positive Results requires Positive Thinking
Hugo Cruz López
Occasional Contributor

Re: Informix on HP

Hello, the shared memory in informix is diveded in: resident portion, virtual portion and the message portion. The biggest portion is the resident portion and the bigger size the best for your system, the virtual portion grows and shrinks constantly. Informix recommends that 75% of the RAM must be assigned to the resident portion,but however yo need to monitor the performance of your system. To grow the size of the resident portion you have increase the parameter BUFFERS in your configuration file and for the virtual portion you have to increase the SHMVIRTSIZE parameter in the configuration file. The configuration file is located in the directory: $INFORMIXDIR/etc/ and the file name is the value of the environment variable $ONCONFIG. But remember that the tunning of a server depends on the constantly monitoring.
I hope this helps.
Michael Tully
Honored Contributor

Re: Informix on HP


We run Informix on a variety of servers, also using raw partitions. We have reviewed our kernel parameters especially in relation to buffer cache. This has actually improved performance substantially. This was occurring on servers that had smallish instances. Attached I've listed some parameters that
we use on an L3000, Informix, raw partitons. I
hope it of some benefit to you.

These days having 2 or 3 times the amount memory as swap is not usually the case. If you had say 8Gb of RAM you would definitely not have 16Gb of swap. This would cause performance degradation.


Anyone for a Mutiny ?
Praveen Bezawada
Respected Contributor

Re: Informix on HP

Hi Tully
Thanks for your information.
Can you give some details about the server configuration on which you have the attached setting, like CPU's, Memory
we are using L2000,4CPu and 2GB RAM
Steve Lewis
Honored Contributor

Re: Informix on HP

I have been in this situation before quite a few times. Java (and Java programmers) accuse the database of going slowly and want it tuned.

What I say is, check the timing of the SQL statements from the application against the amount of time they take to run in dbaccess. If it is quick in dbaccess then informix will be just as quick sending the results back to Java, so Java is the problem, not the DB.

If you are client/server using Informix ODBC driver then I suggest you try a different one, such as SQLlink, because the informix driver is comparatively slow.

With 2Gb of RAM your database will run quickly with BUFFERS at 400000 and SHMVIRTSIZE the same. It does depend on the size of the database (see onstat -p output for read cache % and make sure it is > 99%) and number of users. With raw storage, reduce kernel param dbc_max_pct to 10.
I don't know how experienced at tuning you are because the onconfig parameters you need vary wildly depending on the type of DB you have. What I will say for a start is that
'onstat -g seg' should show just one V segment. If there are more than one 'V' line then increase SHMVIRTSIZE to be the sum of all segments. They will also be listed in
ipcs -ma.