4971RE: PeopleSoft DBA Forum Performance Managment
- Jul 4, 2013
I have to say that just 8 cores across to app server nodes for 3000 users sounds a bit light. Obviously you will need to do some load testing. Not all those 3000 will be concurrently active, but I would think you could easily have 30 concurrent requests.
If you have 3000 users concurrently logged into PeopleSoft that is 3000 threads in the webserver. So you may need several web servers on each physical middleware node and load balance across them.
The important thing is to have enough memory across the various Java pools to support the users. However, I would keep the Java pools down to a reasonable size 1-2Gb each, and configure as many web servers as necessary. I don’t think Physical memory will be your constraint.
With just 4 cores on each app server, you might be able to support 8 PSAPPSRV processes (ie 2 per CPU core) in the application server on each node before the app server node runs out of CPU. There is no point configuring more PSAPPSRVs than you have CPU to service. If the 3000 users make more requests than those PSAPPSRV process can handle they will simply have to queue – that is what Tuxedo is designed to do. That will degrade response time, but the system will run without crashing. This works well when occasion spikes in the load exceed the application server capacity. However, if the load consistently consumes all PSAPPSRVs the queues and time spent queuing just lengthen (30 concurrent requests on 16 PSAPPSRVs on 8 core could get you into this position). Time spent queuing in Tuxedo is a part of the service time, when the time spent queuing exceeds the service timeout the users will experience errors.
The mistakes that I often see are
i) On see queuing in the app server, administrators assume that they must increase the number of PSAPPSRVs, but if you then run out of CPU you just move the queue out of Tuxedo and onto the Unix run queue.
ii) As queue times increase and users experience timeout errors, administrators increase timeout. That is nearly always the wrong response.
In Campus, you might want to segregate students (accessing via the public net) from staff (accessing via internal network) by giving them different URLs to different web servers. And then you might want different app servers behind that. Then you can use OS prioritisation (Tuxedo –n option) to prioritise app server processes. The prioritisation only kicks in when there is no idle CPU.
Where are the process schedulers going? On the same boxes? You might want to run the process schedulers at lower OS priority so that on-line users get CPU in preference.
tel: +44 (0)7771 760660
no its physical server if you have any experience kindly share with me.
That's going to be a challenge. Can you carve out virtual servers? peace,clark 'the dragon' willis
I am going to live People soft campus mangment and Financial modules.My database is oracle 11g R2 on SAN and two application servers with load balancer.No of concurrent users is 3000.Servers configuration is HP DL580P Quad core with 60 GB RAM.My people Tools is 8.52.I just affarid servers is not crashed at peak time.If any one have any suggestions how I can manage 3000 concurrent users load on this hardware I will appericated.
- << Previous post in topic Next post in topic >>