0 Replies Latest reply on Dec 11, 2012 2:32 PM by JunkertF

    Performance metering or How to size configure our webserver architecture?

      Hello people,

      I am in intresting situation, i think actually challange here also the peoples.

      Have currently the following HW's 3 pcs T4-1 in SunCluster (3.2) environment, 32GB RAM per server (64Bit iPlanet Instance).

      for configuring the webservers i tried to use the following guides and benchmarks:
      [SPECWeb T3-2|http://www.spec.org/web2005/results/res2010q4/web2005-20101130-00149.html]
      [MEENA's Blog|https://blogs.oracle.com/meena/]
      My 4 years experience with T1000 and T2000 servers

      That T1000 and T2000 servers configuration (32 bit iPlanet Instance) is depended on
      [Jyri's guide to T series servers|http://www.oracle.com/technetwork/systems/articles/web-server-t1-jsp-141041.html]
      [SpecWeb T2000|http://www.spec.org/web2005/results/res2005q4/web2005-20051205-00018.html]

      The Middle Tier is WebLogic 10 what communicates with the webservers via the weblogic proxy plugin. We have cca. million user per month online in a banking environment.

      I tried to scale the webserver performance with the following tools:
      - Monitoring the server performance with .perf output statistic
      - monitor the ssl key-generation count (T1000 kstat; T4 dtrace)
      - monitor memory usage with mdb

      So i have currently a "working" webserver cluster. I know that the SpecWeb benchmark is not an "all-fitting-clothes", and also there is some issues.

      The issue:
      If we have in the weblogic tier a bit longer running GC period (and somtimes we have on many server paralell), then on the webserver cluster the active session count also begin raising. (Currently we have arround 1400 active session enabled) The problem is when the active session count where be fullfilled and the connections goes to Connection Queue. The Connection Queue size is 81920. But i experience that the webserver can operate only until 10-12000 Queue-ed connections - because the weblogic tier is unable to serve the requests. The question is, that can i more connection handle if i raising the ammount of memory in the server for example to 128GB? Or what can be a good direction to be, to can handle that situation?

      The second question is that can someone give me a good starting point to read how i more precisious performance tuning doing can, or how to monitor, what to check? Is any good document, book, link to that direction?

      Thank You for all the helping (RTFM) "hands", regards,