Re: XCFAS overhead
"How adding new images would negatively impact performance on existing ones even if new images do not share files or databases with existing ones? " As a wild guess: Check the size of the signalling structures, if they support full signalling connectivity and if they follow the size recommendations. (at least one big and one small, and the default set to the big one; preferably no transport classes assigned to any one XCF group, but rather let XCF decide what to do.) "Can isglock structure be split over several cfs - rmf shows we have 4 cfs active - and can traffic be directed to a specific cf? (I'm almost sure that the answer is NO but I might perfectly be wrong)" Scott has already answered that. In addition, check if any one CF has shared processors. If so, insist that ISGLOCK is put into a CF with a dedicated cp. Also check the CF delays, all subchannel busy and path busy. One is caused by lack of physical CF links, the other by lack of logical links. Regards, Barbara -- GMX FreeMail: 1 GB Postfach, 5 E-Mail-Adressen, 10 Free SMS. Alle Infos und kostenlose Anmeldung: http://www.gmx.net/de/go/freemail -- 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
Re: XCFAS overhead
On Fri, 14 Sep 2007 08:04:38 -0300, Walter Trovijo Jr (UOL) <[EMAIL PROTECTED]> wrote: -snip- >And my questions are: > >-How adding new images would negatively impact performance on existing ones even if new images do not share files or databases >with existing ones? In a sysplex, there are shared resources used by the various sysplex-wide components (e.g. ISGLOCK structure, CFs, couple datasets, security DB) that, if held up by a particular member, can cause slowdowns across the sysplex (a/k/a sympathy sickness). You probably have to 'drill down' to see what XCF resource is involved in the delays reported by RMF. >-Can isglock structure be split over several cfs - rmf shows we have 4 cfs active - and can traffic be directed to a specific cf? >(I'm almost sure that the answer is NO but I might perfectly be wrong) No. ISGLOCK must reside on a single CF. Of course, you can set the preference list for ISGLOCK to be a particular subset of CFs and rebuild the structure to get it to the preferred CF. No outage is required. Scott Fagen Principal Architect Enterprise Systems Management -- 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
XCFAS overhead
Hello everybody, I've seen some weird XCFAS behavior in one of our systems recently that I'm trying to understand/explain so I need your help. This used to be a 3 way sysplex on 2 cpcs and everything was fine. The company I work for then decided to outsource it and we kind of lost control over what happens and sometimes we get some not-so-nice surprises. This started to happen on our development image last week; it's a 300mip capped system, so it normally runs at 100% lpar cpu, with a dozen of db2 ddf threads, batch jobs and tso regions. But since then, with appearly the same loads it hungs cpu constrained. Rmf mon3 showed xcfas as the primary cause of cpu wait for every application. Looking further into rmf mon3, I uncovered that isglock access rate increased by 10 times when system was slow; I also found that our sysplex turned to 11 systems with at least 6 of them with production kind of load. My conclusions are: -The capped lpar is suffering and loosing competition on sysplex resources due to it's low priority. -Some hardware/software configuration change is making this (and maybe every) image to make a lot more grs calls. And my questions are: -How adding new images would negatively impact performance on existing ones even if new images do not share files or databases with existing ones? -Can isglock structure be split over several cfs - rmf shows we have 4 cfs active - and can traffic be directed to a specific cf? (I'm almost sure that the answer is NO but I might perfectly be wrong) Thanks a lot in advance for your help, Walter Trovijo Jr. -- 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