Hi all,
We have questions concerning searches in Enterprise Portal 6.0. We originally installed EP6.0 and TREX on the same UNIX (Solaris) host. We recently ran some load tests which indicated our installation was inadequate to support the projected usage. What we experienced is that performance navigating from role to role, workset to workset and page to page, while not stellar, was acceptable. We then started including searches in the load mix, the jlaunch processes rapidly climbed, consuming 40 - 50% of the CPUs, throughput went down, and response time went up. It got so bad that users could not log on and searches would time out or never complete.
So, we installed TREX on a separate host. Once again during load testing, navigational performance was acceptable until we started loading searches into the mix. Again, on the portal host, the jlaunch processes went to 40 - 50%. This time however, searches did not seem to be impacted and completed as one might expect them to. The TREX host hardly showed any change in processing statistics before, during or after the load test.
It appears to us that the TREX processes are not much of a load on the TREX host, but the actual search requests are an extreme load on the portal host which significantly impact it's performance. Is this normal and/or expected behavior? These tests were done with about 25% of our expected load or about 200 - 250 users.