Daniel, I don't have experience with WAS. But I think that I once heard that WAS has some java in it. With Java, you have Java Heaps.
We are a SAP shop. SAP's sizing recommendations for heaps start at 2GB. By the time we allow memory for other stuff the linux system needs, it is real easy to have a linux guest sitting at 3GB or so. On a slightly related topic. We have a non-SAP webserver running tomcat on SLES10 SP3 Intel hardware. Due to a combination of factors, it is running with a 5GB heap. The 5GB heap size comes from a combination of how the developers wrote the application, the garbage collection parameters that the developers insist on using, and some "third party code" that is running on that Linux system. All this to say, in my somewhat limited experience with Java, proper heap sizing along with appropriate garbage collection parameters can have a dramatic effect on performance. Do you all already have experience with heap sizing and garbage collection tuning? You mention, you want to know about the storage required for WAS systems. You did not mention IFLs. Getting back to garbage collection tuning, some of the garbage collection methods can "Stop the world" for varying amounts of time. If your applications are written in such a manner as to require a large Java heap, you have to figure out how to minimize the amount of time that the world is stopped. (You don't want your WAS system to go away for 30 to 40 seconds periodically do you?) Sometimes you have to define an extra processor so that garbage collection will finish in a timely manner. We start our our SAP j2ee systems with 2 IFLS. We have one system running under VMWARE that we have had to assign 4CPUs to in order for garbage collection not to be too noticeable. On a slightly related topic, I noticed that you were running SLES11 GM. Have you thought about SLES11 SP1. There was a webinar a few weeks back. The whole gist of the presentation was letting people know how much performance work had been done on SLES11 GA to improve performance. (It seems that in several areas that SLES11 GA is actually slower that SLES10 SP3). There might be something in SP1 that will help your situation. I know this is not WAS, but... As far as sizing of Linux guests go, we have one running the SAP BI/BW module that is 18GB in size. AND IT NEEDS TO BE BIGGER. Right now there are times when this linux guest is experiencing 5 digit Linux swap rates. The developers are working on things to reduce storage consumption. I googled Webshere application server sizing and came up with the following. http://www.ibm.com/developerworks/websphere/techjournal/0909_blythe/0909_blythe.html Looking further, I see these sentences.
Basic tuning: <http://www.ibm.com/developerworks/websphere/techjournal/0909_blythe/0909_blythe.html#sec3>This group of tuning parameters covers a few of the most commonly tuned and heavily used application server components, starting with the JVM. These settings traditionally provide the most bang for the buck.
The first item under basic tuning is JVM Heap Sizing. WAS appears to be the same as SAP. You have to size your heap properly. And the way that the developers wrote the application will partially determine the garbage collection method. Which can influence your heap size. Which will help to determine your IFL allocation.... Hope this helps, Ron On 9/15/2010 5:25 PM, Daniel Tate wrote:
In a day or so, I will be asking for sizing advice (practical, not redbook - redbooks and other theoretical sources have not been too beneficial to us - this mailing list on the other hand has).. but I would also like to know of any good resourcesdy (message boards, mailing lists, fidonet feeds, whatever) for websphere, especially websphere on Z.. We've given up on using LDAP SSO because of the memory consumption that nobody was able to find the source of (LDAP SSO turned on = uncontrolled memory growth until OOM_KILL even without apps loaded onto the appserver.. turn LDAP off, no problems).. Now we're onto sizing. We only have a dev environment with 7 servers, the LPAR has 15GB of storage, and we know we are going to need more. We will be looking for not only a number but verbose justification for that number (executives will demand that).. so we need as much ammo - theoretical and practical experience - as we can possibly gather. Our mainframe admin was expecting 15GB to be plenty, as we run our entire primary application (z/OS) on less memory than that - and he fully expected 1-2GB per system to be MORE than enough). I have asked our WAS admin for details of the final planned environment to pass along to as many people as I can to get real world advice. So does anyone have any good resources? This is WAS 6, SLES 11 GM, z/VM 5, on a z9. ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/ .
---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/
