Charles Harbour <charles.harbour@PEARSON.COM> posted for Gunnar: > The p550 is running the new Advanced POWER™ Virtualization with dynamic > LPAR and micropartioning managed from a HMC. All this was setup by a > consultant. At the moment there are only 2 partions, the virtual i/o server > and the main running partion. This main partion has right to use all > available resources minus what the i/o server uses. I have 100% control of > what's running except for what AIX and the i/o server does behind the > curtain, so no other users. memsize = 2GB, bufno=default=1. SASWORK is on a > FastT900 SAN-system. Yes, I'm using fullstimer, this is the result from the > both p-series machines: > > p550: > > PROCEDURE GENMOD used (Total process time): > real time 6:21:19.92 > user cpu time 3:52:36.61 > system cpu time 5.54 seconds > Memory 801645k > Page Faults 350 > Page Reclaims 1041561 > Page Swaps 0 > Voluntary Context Switches 457314 > Involuntary Context Switches 1152405 > Block Input Operations 0 > Block Output Operations 0 > > p615: > > PROCEDURE GENMOD used (Total process time): > real time 9:24:59.58 > user cpu time 9:13:53.78 > system cpu time 18.63 seconds > Memory 801645k > Page Faults 858 > Page Reclaims 693320 > Page Swaps 0 > Voluntary Context Switches 676930 > Involuntary Context Switches 2166843 > Block Input Operations 0 > Block Output Operations 0 > > My first thought was also that some disk i/o was going on but we have > carefully monitored the disk with nmon64 and after loading the dataset to > RAM it hardly uses the disk at all, look at the low page faults. We have > also watched what the i/o partion does and it's resource use is very low. > As you can see the p615 has a normal difference between user cpu time and > elapsed time. We also had an issue with proc phreg on the p550, where we > got the same big difference but there it actually had to swap a 40GB > working file in/out, running with the multipass option reduced the real > time with 3 hours and there was a normal difference between user cpu time > and elapsed time.

In both of these, the count on involuntary context switches is massive, and page faults are really low. Could there be a problem with the tuning of the dynamic LPAR ? Could there be a problem with underlying system processes? I'd want an AIX expert to look at these numbers, and I'd want SAS Tech Support to try out your cases on their AIX boxes.

David -- David Cassell, CSC Cassell.David@epa.gov Senior computing specialist