Relationship between Eqalloc and Eqmemsize

[Eric Leung_2]

Hello experts,
There are more that 150 hosts in my site and I am investigating a core dump problem. According to HP based on analysis of the crash dump, the problem was caused by exhaustion of equivalently mapped reserve pool. The remedy is to tune the eqmemsize from 15 (default) to 80. In addition, HP also told me to use "kmeminfo" to monitor the value of Eqalloc.
However, I found that the eqalloc is ever increasing from 68 (after reboot) to 125 ( after about 2 weeks). I wonder if there is still risk that system panic may occur. Actually it does not occur after the eqmemsize was tuned.

I have also got the eqmemsize and eqalloc from my servers ( as attached ). I can only conclude that the value of Eqalloc is generally smaller than eqmemsize on my A and L-class server. However, the D-class counterparts behave differently. The eqalloc value is much larger that the eqmemsize in some servers.

Would some experts can tell me that whether I am facing a problem.

Some servers have a very large eqalloc value, is it normal?

Some servers have error message in dmesg output (as attached) but some do not, what is the criteria that the error message appears in the log?

Can the the error message in dmesg be direct to syslog?

Is Eqalloc is an correct indicator of the usage of equivalently mapped reserve pool?

What is the strategy should I take to tune the eqmemsize value. As each server have different configuration, such as CPU, Memory and disk.

In fact, there are still a lot of questions in my mind. Hope somebody can help.

[T G Manikandan]

http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000063200023

[Eric Leung_2]

Thanks for your information.

However, what is the reason why the eqalloc value is higher in D-class server? Is the number of outstanding, pending I/Os in parallel in D-class is higher than A/L class counterpart because of the performance difference provided both are running same type of application?