Re: onstat -g questions
Posted in 1996
On 22 Apr 1996, Rajiv Jauhari wrote:
> Has anyone used the "onstat -g sch" and "onstat -g spi"
> commands to help tune the performance of Informix 7.1 on an SMP?
> If so, can you explain the exact meanings of the various columns?
I recently made an adjustment based on the info from onstat -g glo and
onstat -g sch. I am running 7.10.UC3X1 on an 8 CPU Sequent.
When a VP (CPU, SHM or TLI) has no work to perform it enters a loop to avoid
being swapped. This is a busy wait.
My production engine has 4 CPUVPs. I have 3 TLI poll threads and 1 SHM
poll thread. The TLI poll threads are run in the CPUVPs. The SHM poll
thread runs in its own VP.
The poll threads always seem to be executing as seen in onstat -g act.
With one more CPUVP than TLI poll threads, the fourth CPUVP would
accumulate 5.6 million busy waits and 3.8 million semops over a week.
The other CPUVPs had under 200 busy waits. The cpu usage of the looping
CPUVP was 2.5 times that of the other CPUVPs.
I also had a seperate process for the SHMVP.
onstat -g sch output: ( not complete )
VP Scheduler Statistics:
vp pid class semops busy waits spins/wait
1 626 cpu 165 203 9056
3 643 cpu 6 13 6963
4 644 cpu 6 7 9592
5 645 cpu 3870117 5627502 8090
10 650 shm 35 129 505
onstat -g glo output: ( not complete )
Individual virtual processors:
vp pid class usercpu syscpu total
1 626 cpu 69804.92 12617.92 82422.84
3 643 cpu 68026.23 11544.72 79570.95
4 644 cpu 62032.02 10149.14 72181.16
5 645 cpu 163929.71 8552.54 172482.25
10 650 shm 33447.84 7598.44 41046.28
tot 397323.76 52474.04 449797.80
What I did was to run all poll threads through the CPUVPs. I wanted
get some use out of my CPUVP that was looping and eliminate the SHMVP.
This is what onstat showed after.
onstat -g sch output: ( not complete )
VP Scheduler Statistics:
vp pid class semops busy waits spins/wait
1 601 cpu 172 266 7848
3 613 cpu 10 12 9377
4 614 cpu 5 7 9343
5 655 cpu 5 9 5644
(no more shmvp)
onstat -g glo output: ( not complete )
Individual virtual processors:
vp pid class usercpu syscpu total
1 601 cpu 70188.90 12097.90 82286.80
3 613 cpu 67164.41 11039.52 78203.93
4 614 cpu 61346.75 10026.08 71372.83
5 615 cpu 131769.77 8184.28 139954.05
tot 330501.19 42149.71 372650.90
The final results are I reduced the total cpu usage and the cpu usage of
the fourth CPUVP and eliminated the seperate SHMVP. I was suprised that
the fourth CPUVP still had so much usage. I guess I resolve that
another day.
Regards,
Steve Berg
spb@masada.nb.com