On Thursday 03 May 2007 12:11, herta.vandeneynde@(protected):
> We run Oracle RAC 10.2.0.3 over ASM on a two member cluster. Each member
> has 64GB memory and runs x86_64 SLES 10 (2.6.16.27-0.6-smp). We use
> EMC's powerpath driver, which doesn't support any higher version yet, so
> we're stuck with this version for now.
> Hugepages are enabled and used:
> $ cat /proc/meminfo
> (...)
> HugePages_Total: 16384
> HugePages_Free:  5425
> HugePages_Rsvd:   20
> Hugepagesize:   2048 kB
>
> This non-production cluster is used by developers, application and
> acceptance testers. Each member runs an Oracle Enterprise Management
> Agent, a Big Brother Agent, and 1 instance of each of the 36 databases.
> The sum of the sga's of 36 database instances is 21 GB, which are taken
> out of the HugePages, and therefore are locked in memory.
> The sum of the aggregate pga targets is 3 GB. I know the pga aggregate
> target is below the recommended 20%, but these are not end-user
> production databases, and the OEM advisor shows current setting to be
> adequate: only a few of the development databases show a handful of
> executions have been non-optimal executions.
>
> Note that the sum of all SGA_TARGETs + PGA_AGGREGATE_TARGETs is less
> than half of available physical memory.
>
> So how come we are swapping? Immediately after reboot, i.e. even before
> we have user activity, about half of our swap space is used (4 GB out of
> 8 GB swap). Following the startup, sar mainly shows swap ins.
>
> FWIIW, a second cluster with the same hardware (apart from faster
> storage) and software setup, has 12 databases defined with a total of 31
> GB of sga and 7 GB page aggregate target. There's no swapping on the
> cluster yet(?) (SwapTotal = SwapFree), but most of the databases are
> still empty (data not
> imported yet).
>
> This second cluster will become our new production cluster as of next
> weekend. Needless to say we're anxious to understand what causes the
> swapping on the non-production cluster.
>
> Any suggestions on how to debug this?
>
> Kind regards,
>
> Herta

Hi,

Just two checks come to mind:
- do your servers have the same swappiness value (/proc/sys/vm/swappiness) ?

- also, if you are using async i/o, see the following article
about /proc/sys/fs/aio-max-nr and /proc/sys/fs/aio-nr:
http://www.novell.com/support/search.do?cmd=displayKC&docType=kc&externalId=3445451&sliceId=SAL_Public&dialogID=34533676&stateId=1%200%202384069

(To compare the kernel settings in one step, you can compare the output
of "sysctl -a" from all nodes .)

HTH,
Didier

--
To unsubscribe, email: suse-oracle-unsubscribe@(protected)
For additional commands, email: suse-oracle-help@(protected)
Please see http://www.suse.com/oracle/ before posting