rmap overflow and file table is full in syslog.log

[Tiem Nguyen Van]

Hi all,

We are using 2 node clusters on HP rp5470, share VA7000.
Currently, it has problem with performance.

In the syslog there are some error:
...
vmunix: quad4map_32bit: rmap ovflo, lost [863274,863283)
vmunix: file: table is full
....

Could anyone show me what the log mean and how to correct the problem ?
Best regards,

Tiem NV

[Pete Randall]

Not sure about the other one, but the "vmunix: file: table is full" is caused by the kernel paramter nfiles being set too low. Go into SAM and increase nfiles and build a new kernel. That will fix that one.


Pete

[Joseph Loo]

hi,

as pete has mentioned, nfile is low which resulted in the second error message.

for the first error, refer to these doc:

http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000063201277

http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000063200449

it has something to do with needing to increase another kernel parameter, shmmni.

regards.

[Simon Hargrave]

The file table is full error is caused by your system trying to open more files than the kernel will allow, as defined by the kernel parameter nfile.

YOu can monitor the status of your files table. If you have glance, you can view it there. Alternatively use:

sar -v 10 10

this will show a column file-sz, which will show current amount of files open and maximum allowed. Your tuning should be based on this. Knowledge of the applications you are running on the server (eg regularly opens new files, or opens lots and leaves them open like oracle) will also help.

[Bill Hassell]

For the error:

> vmunix: quad4map_32bit: rmap ovflo, ...

you need to increase shmmni to give more room for the quadrant 4 resource map. It is caused by excessive fragmentation by 32bit programs. This fragmentation can be caused by a lot of stopping and restarting of processes that use shared memory, and especially if processes are terminated with kill -9, an especially bad thing for databases. kill -9 leaves orphaned shared memory segments (ipcs -bmop) and related files incomplete.