Re: [suse-oracle] Huge Pages and SAP

Werner Flamme <[email protected]> Wed, 20 Jul 2011 19:40:38 +0200
Newsgroups gmane.linux.suse.oracle.general
Message-ID <[email protected]>
Austin Joseph [23.04.2010 19:26]:
> 
> Hi List , 
> 
> One of our customer after some convincing has configured Huge Pages on their clustered two node SAP , Oracle server. But it has been observed post implementation of Huge Pages on Oracle system as expected the memory utilisation has become efficient and memory utilisation has dropped to 65 % for the same workload from 95 % which was the case before Huge Pages Implementation. Oracle version is 10gR2 and memory is 128GB  
> 
> But on SAP node which has similar Huge Pages setup being a clustered environment , the percentage memory used has shot up to 60 % from ~ 21% which was the case prior to Huge Pages implementation on SAP node. The SAP version is 4.6c and SAP certainly does not support Huge Pages. 
> 
> So can this shoot up of memory % can be seen as a normal behavior for SAP ? If so from other customer experience is it desirable to disable Huge Pages on SAP node while keep  continuing  using Huge Pages for Oracle node ? Since this is a clustered environment it was thought to be desirable to configure Huge Pages on both nodes. 
> 
> Thanks & Regards 
> Austin 

Hi Austin,

we have several SAP systems (ERP 6 EhP 4, SolMan 7 EhP 1) running on
SLES 11 SP1. We introduced huge pages due to us sharing the OCFS2
volumes between the SAP and the database cluster nodes. We did not see
any increase of memory usage, but instead a decrease. Maybe
coincidentally with the upgrade from ERP 2004.

The RAC runs with Oracle Linux 5, Clusterware 11.1, DBs are 10.2.0.4.

We used the calculating script provided by Oracle to get the number of
hugepages on the various VMs. They were always different numbers, so we
used the highest number to set the value on all SAP boxes. The new
setting worked after the first reboot, because on the running systems
there was not enough free memory to reserve the hugepages at once.

We gained an enormously higher reliability of the OCFS2 system, which
tried to reboot several nodes whenever heavy load was encountered: we
had no OCFS2 related crashes ever since we introduced the hugepages.

HTH
Werner
_______________________________________________
suse-oracle mailing list
[email protected]
http://listx.novell.com/mailman/listinfo/suse-oracle