The following is from the Performance Toolkit from z/VM 5.1. It shows a
difference between TCPU and VCPU as 3.66 seconds, which I assume is the CP
overhead. I watched this for a few minutes this morning and I can see the
difference stays between 3 and 4 seconds.
Userid %CPU TCPU VCPU Ratio Total DASD Avoid Diag98 UR Pg/s User
Status
>System< 1.14 .681 .631 1.1 4.9 4.8 .0 .0 .0
> .0 ---,---,---
ZOSONE 84.3 50.60 46.94 1.1 368 364 .0 .0 .0 .0
EME,CL3,DIS
ZOSTWO 4.49 2.695 2.561 1.1 3.7 3.3 .0 .0 .0 .0
EME,CL3,DIS
After I copied the above I did a IND USER ZOSONE and notice the difference
between VTIME and TTIME.
USERID=ZOSONE MACH=ESA STOR=1000M VIRT=V XSTORE=NONE
IPLSYS=DEV 0764 DEVNUM=00265
PAGES: RES=00255997 WS=00255997 LOCK=00000003 RESVD=00000000
NPREF=00000000 PREF=00000000 READS=00000001 WRITES=00002553
XSTORE=000003 READS=000071 WRITES=000074 MIGRATES=000000
CPU 00: CTIME=06:38 VTIME=088:05 TTIME=274:56 IO=296239
RDR=000000 PRT=006680 PCH=000000
Is VTIME the VM overhead in minutes and seconds? Is that included in
TTIME? If so the overhead of VM from the IND USER is over 30%. The
ZOSONE machine was last IPL'd (logged on) on Wednesday October 12th around
9 PM.
No, VTIME is virtual (emulation mode) CPU time, corresponding to the
PerfKit's VCPU value, and TTIME is the total CPU time, corresponding to
PerfKit's TCPU value. And yes, VTIME is included in TTIME, but the CP
overhead is the difference TTIME-VTIME. Note also that the IND USER output
doesn't show the 'thousands' digits for CPU minutes: the fields may have
wrapped.
If you issue several IND USER ZOSONE commands in the same interval that is
used for collecting CP monitor data (default is 60 seconds) the deltas of
their output for VTIME and TTIME should be similar to the VCPU and TCPU
values displayed by the Performance Toolkit.
Eginhard Jaeger
