Hermann L. Huebler commented on Jul 25, 2016

Re: IBM Digital Experience: WebSphere Portal V8.5 Performance Tuning Guide

When preparing Linux for HugePages be aware that at least RedHat Linux supports the concept of anonymous. This means that other applications (for example the nodeagent) might reserve HugePages although these are not configured. The consequence is that not sufficient HugePages are left for the Portal Servers Heap and therfore normal pages are used. To address this and make sure that HugePages re used make sure that:

a) the number of "HugePages_Free" BEFORE starting the Portal server is sufficiently high to hold the full Java Heap

b) verify after the Portal server is started that huge pages are used. This can be done by checking the pageSize attribute in the initialized section of the verboseGC output. This attribute should look like: when 2MB pages are used.

Stan Logan commented on Feb 3, 2015

Re: IBM WebSphere Portal V 8.5 Performance Tuning Guide

The directions for disabling JCR Text Search still lists the icm.properties change as well as the Resource Environment Provider setting.

Also, there is a known memory leak condition with Apache's HTTP Server when MaxRequestsPerChild is set to zero.

(per apache.org's documentation: "Setting MaxRequestsPerChild to a non-zero value limits the amount of memory that process can consume by (accidental) memory leakage.")

Hunter K Presnall commented on Sep 26, 2014

Re: IBM WebSphere Portal V 8.5 Performance Tuning Guide

For version 1.2, I removed the start order statement in the SharedClassCache section and mention that the DMGR and node agent should be changed as well.

Gabriel NKUITE commented on Sep 10, 2014

Re: IBM WebSphere Portal V 8.5 Performance Tuning Guide

On page 12 you said "After making the change for Portal, it must be restarted first to avoid the DMGR or node agent’s default setting being used. Another option is to also make the cache size setting on the DMGR and node agent in addition to Portal so that start order is not critical."

Because portal in a cluster cannot be started before nodeagent, I think it will be better to only recommend to make the cache size setting on DMGR and nodeagent to avoid any trouble.