We are going into production, tomorrow, with DB2 Connect Server V9.5 fixpack 2. A Cashier system, with about 8 users.
I'm on 160 MB virtual, 1 3390-9 drive, two vdisk swap files (about 10,000 4K swap pages used over 3 weeks). This is on a z/890, with the IFL about 20% busy. And no VM paging. Only problem we have had, and can't seem to shoot, yet, is we are using stored procedures on DB2/VSE. About 1 in 10 transactions, has a 20-30 second startup time. Not the first, not the last. Just blind sided. Every box in the transaction path, was not having a CPU problem, an IO problem or a Paging problem, when we had poor response time. Did you have that, and did you solve it? On the DB2 side, my guess is you have too much virtual storage defined. You should have automatic memory management enabled (by default). So scaling back will also scale back the DB2 cache. Check the hit ratio. But I wouldn't try to keep the hit ratio at the same level as the other platforms. If you are on ficon, you have one hell of a kick ass I/O subsystem. Trade some memory for I/O, within reason. That gives you more memory for more servers, until you hit CPU limitations. BTW, my test UDB servers are 512 MB each. Only 2 3390-9 packs for data. Yep, the first time thru, the application ends up loading DB2 cache, and then the queries fly. Testing doesn't need a lot of cache. I don't have Websphere, so I can't consider it's memory loads. Tom Duerbusch THD Consulting >>> <[email protected]> 5/19/2009 11:13 AM >>> I'm finally in the process of firing up some "boxes" under VM to replace some aging AIX critters and I could use some advise as to real memory and swap allocations for these new guys. 1st victim was a DB2 connect enterprise unlimited server - cut it to production 2 weeks ago and everything is good. 246900k of memory and 3G of swap, so far, so good except for a small issue with java versioning v1.5 on new box, v1.3 on oldest client app - but that is a java issue that would show up on any platform with the new version. 2nd victim is a DB2 ESE and WebSphere MQ server - can't even fire up the tools database because the bufferpools don't have enough memory. This is where the advise is needed since the IBM docs don't have a recommended starting point for the z/VM environment ( DB2 9.5 fixpack 3b) and I really don't want to scarf all of the vm resources for this machine when the grand plan ( insert evil laugh here ) is to put as many AIX and intel linux onto this platform (Z) as we can and still maintain decent performance. Current AIX box is a 2-way power3 box with 16G of memory - low utilization in a development environment. 3rd victim is a WebSphere AppServer box - 6-way power4 box with 16G memory. Low to moderate utilization in a development environment with a second instance of WAS for background production use ( the moderate utilization part - when it is active ). Current VM lpar has 2 IFL, 9G cstor, 3G estor, z/VM 5.3 and SuSE 10.2 guests Thanks, Bruce ---------------------------------------------------------------------- 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 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
