On Sun, 15 Aug 2010 12:23:47 -0700, Gerhard Adam <[email protected]> wrote:
>> We are still using large pages but have been doing it in steps over the >> last 6 months (IPL with larger LFAREA, convert more WAS regions to 64-bit). > >OK, I give up. Why? What is the benefit versus the cost? Even the original literature suggested that it could cause performance degradation for some applications (although obviously not because of bugs). > There is no cost. It's just a matter of IPLing with LFAREA. Right now because of the "bug" I wrote about in a previous post, it just takes a lot more planning than it should have. The only application on our WebSphere LPARs are WebSphere (and DB2 via datasharing). >So I'm curious. Who is actually measuring this? What is being measured? and how did anyone determine that it would be beneficial? > It has been measured and IBM has given some presentations about it at SHARE and I assume zExpo. I'm not sure where to find it in black and white at the moment (maybe someone else does) and the SHARE web site for proceedings appears to be down. Maybe there have been some CMG papers as well. These saving are in memory management (less TLB misses). In addition to memory management savings from z/OS, I think there cab be savings in Java from less GC (garbage collection) since the JVM heap size can now be much bigger than the 512K or 1M people might have used without 64-bit Java. Changes were made to z/OS to allow the "hole" between 2G and 4G to be used in order to accommodate the changes made to Java in order to achieve the benefit. Actually, I think these changes were made just to make 64-bit Java perform on an equal level with 31-bit Java. Otherwise the large pages alone wouldn't have been enough to offset the performance degradation of 64-bit Java DAT. That is why it is recommended to stick with the 31-bit flavor for most applications. The performance team at my client keeps a close eye on this sort of thing, and I assume they have measured the benefit in MIPS vs # of transactions before and after the changes or they wouldn't go through the migration. On another note... sometimes you just have to take IBM's word for certain recommendations as they are very difficult to measure outside a lab. One example that comes quickly to mind is CPENABLE. Regards, Mark -- Mark Zelden - Zelden Consulting Services - z/OS, OS/390 and MVS mailto:[email protected] Mark's MVS Utilities: http://home.flash.net/~mzelden/mvsutil.html Systems Programming expert at http://expertanswercenter.techtarget.com/ ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
