Saturday, September 08, 2007

Concurrent Usage for E-Business

Blogger seems to have lost the last entry that I posted earlier this week. I didn't check and now I can't even remembered what it was that I posted about.

Anyhow, there was a number of stuff to post but I thought I would focus on a current technology issue that we are facing with our Oracle E-Business project. Our test lab has been running some performance scripts on our E-Business test environment and found that it was lacking. For example, one of the test was to see how many concurrent users the environment can support. According to the results, the environment maxed out at 80, 81 concurrent users with the systems statistics showing that all four CPU were at 100% max and subsequent attempts to increase the users resulted in failures (i.e. Error 404 - Page Not Found). Now, I'm perplexed as our initial sizing done with Oracle and HP showed that our old server (application tier) was able to support 200 users (i.e. 2 CPU x 650MHz with 16Gb RAM - PA-RISC processors) and our new server (application tier) is a 4-CPU x 1.6GHz with 32Gb RAM - Intel processors). My expectation was that we should be able to support 100 users without any problems but the results showed only 80 (at the very max). Now I understand that it's an apples and oranges comparison but surely a 4-CPU x 1.6GHz Intel Processor can't be slower than a 2-CPUx650MHz PA-RISC Processor or were there something else afoot that I am unable to grasp. Obviously there is a disconnect somewhere and I need to find that before we can say that our current hardware for the E-Business environment is unable to support our internal usage much less allow us to expand in the future.

This is causing a fair amount of concern for us as we are scheduled to go live with the Order Management modules within the next four weeks. For those readers who has E-Business Suite at their place of work, what are your hardware specifications for your application tier? We are debating whether we will be able to scale horizontally by adding another node (application) and a load balancer to the environment and have everything ready (tested) within 4 weeks.


Anonymous said...

You should check latch waits. You are probably running into some issue where the faster cpu allows more updates to the SGA, perhaps some hot header block, so the latches start chaining up and spin the cpu.

Anonymous said...

Thanks, I will pass it onto the DBA group to take a look at.