brankley@pdn.paradyne.com (Bob Brankley) (10/04/90)
I have been running UTS 2.0.1 since March and have been very happy with it. I have had a couple of small problems with UTS 2.0.1, but the problems were not major and the UTS Support Center has been very helpful in resolving them. One of the current problems I am pursuing concerns excessive paging under high CPU use. The UTS Support Center is currently working on the problem, but I was wondering if anybody had seen this problem before. Here is my system configuration: System configuration: Dual processor IBM 4381 with 16 MB of main memory, running VM/XA 2.1. UTS runs as a guest under VM and is isolated to run only on the second processor. UTS is configured to be a 16 MB 370-type virtual machine. Our disks are IBM 3380 DASD. The system is rarely in use by more than three users. Problem: According to the output of "sar," the number of real page faults increases dramatically as UTS's CPU utilization approaches 70% or above. These are real page faults generated by the UTS paging algorithms and NOT VM/XA. The number of virtual page faults (those caused by VM paging out portions of UTS's pseudo-real storage) is extremely low during the same period. Another caveat is that UTS is paging despite the fact that he has yet to fill what he perceives to be real storage. In fact he has typically only consumed 9 MB of the 16 MB of VM virtual storage he is allocated. I have isolated one application which seems to reliably cause the above anomaly. It is a makefile which uses vmpunch to assemble a set of source files on an MVS guest machine running on the other processor. Nightly accounting also causes heavy paging activity. With a processor load between 70-100%, the number of real page faults easily reaches 40-60 rpflt/s, rendering the virtual machine unusable. The processor time used by the wss_daemon goes through the roof as well. Has anybody else observed this activity in UTS 2.0.1? Could this be a local mis-configuration issue which might easily be solved by applying a few sticky bits? Any other ideas? I look forward to any discussion on this topic. Bob Brankley System Analyst AT&T Paradyne brankley@pdn.paradyne.com