Performance on N-class + Oracle 8.1.6.2 + ServiceGuard

[Geert Buelens_1]

Hi,

At two customer sites we are experiencing performance problems with Oracle 8.1.6.2. (even before upgrading from Oracle 8.1.5) running on N-class machines with ServiceGuard.
Our application is running very well on other HP type machines (Oracle 8.1.6.2+ServiceGuard are configured the same) : R380, A500 & A180

Our application is written in Java and uses JDK 1.1.8.04 + weblogic app server 4.5.1. It exists out of multiple programs which are run by ServiceGuard.
CPU usage when processing an input file (put the input date of the file in the database) is dropping to to 5-10% when we are running more and more packages. There is no CPU, memory or I/O problem. It looks like the system is not scheduling processes that are ready to run.

To proove it is not caused somewhere in our application, we found also that with only the database running, an export takes 13 hours if we run the command on the same machine as were the database is running and only 15 minutes if it is run from the other node.

On Oracle forums (metalink) we found already two customers with (similar?) performance problems when using an N-class, Oracle 8.1.6 and ServiceGuard. An HP consultant mentioned a third case of which he would try to find more detailed information for.

Is anybody having simular performance problems with N-class machines, Oracle and ServiceGuard? And what cwas or could be the solution?

Thanks already in advance for your interest,

Geert

[Peter Jansen]

Hello Geert,

You should really get somebody who will take a deep look at the system for bottlenecks on the resources (CPU, Memory, Disk IO, LAN), also with the export running. This can be done using the standard Glance/MeasureWare tools. Also, a check of the system configuration and Oracle configuration should be done.

Sorry I do not have a direct solution for your problem, but it seems to me that a deeper investigation needs to be done.

Best regards,

Peter Jansen
ITchallenge

[Ed Ulfers]

I had a friend with a similar issue, it turned out to be the system overhead associated with routing (re-routing) traffic through a virtual network interface. The routing took a surprisingly high amount of CPU resources, try checking CPU resources at the time of slow performance.

As far as Oracle and an application running on the same box, check into using the Bequeath adapter via tnsnames.ora (the difference of communicating through memory versus layering through the OS and networking aspects is astounding with large data volume or frequent new connections).

[Carlos Fernandez Riera · Accepted solution]

Yes ... I know.

I had reinstaled a V2200 (5 Gb , Oracle 7 , Oracle8 ..) on Feb. 2000 w/

XSWGR1100 B.11.00.47.07 General Release Patches, November 1999 (ACE)

and update to path level Feb. 2000.

Our dayly exports need each day more and more time, beggining in 1h and after few days 5h. After several weeks following this we apply:
PHKL_21610 1.0 Large Data Space, kernel memory leak fix
( and others of course).
Now all is right.

[Geert Buelens_1]

Hi,

Something that would be interesting to know is that the installation of HP-UX was done early in 1999. It was one of the first N-classes that were shipped. Since then we applied patch bundles and some individual patches (done by HP consultants), but it did not solve the performance problem.

I personally think there must be a difference in an old kernel with a patch bundle and a new kernel that is shipped today. I'm not sure about it, but I can imagine HP is not shipping any kernel update as a patch. Is this the case?

As a next step I asked for a completely new HP-UX installation (this will take us some weekends) on new boot disks which will be swapped with the current boot disks during this upgrade period.

It would be nice to know in advance if anybody has a well performing combination N-class, Oracle and ServiceGuard?

Thanks,

Geert

[Dieter Degrendele_1]

Hi,

We are using a 2 node cluster with N-Class servers, Oracle 8.0.6.1 and MC Service Guard. We did not have any performance problems, but I have to confirm that the network speed was very low in the beginning. We changed the the primary network cards from 100BaseT to Monofiber and since then the speed is Ok.

Regards,
DD