[HPADM] RE: SUMMARY 8420 VPARs running with High load on CPU0 and low load on all other CPUS

• Previous message: Jeff Lightner: "[HPADM] rw as part of hostname for NFS export"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Date: Mon, 21 Mar 2005 11:45:12 -0500
To: <hpux-admin@dutchworks.nl>

I wanted to thank Bill Hassel and Ted Fisher for their assistance on
this issue.

Bill provided some wonderful insight into the noted issue:

Not at all unusual. There are likely dozens of processes performing some

sort of poll or signal check, then going to sleep for a short time. The

runqueue (poorly named: load in uptime) is really the number of
processes

waiting for a CPU. The queue assigns the first available CPU (probably

the second one, the process runs then goes to sleep. top and uptime

are very poor in representing short-lived proceeses. If this were truly
a

CPU load, both CPUs would be about equal. but the crude metrics in

sar and top can't account for processes that last only a few
milliseconds.

So this has nothing to do with vpars. It's an artifact from short-run
processes.

Since the users see no impact, all is well. I've seen 400 polling
processes

create runqueues of 40-50 on a 2 processor system with no impact on

response time.

   -Jim

Original Message:

I have an 8420 with 2 npars. Each npar has 2 vpars.

What I have noticed once I put Oracle Apps on one of the vpars (2 CPU,
16 GB ram), is that CPU 0 has a load of 20 and CPU 1 has a load of 0-5.
CPU Utilization is evenly distributed over both CPUs. I have noted the
same behavior on most of the active vpars in use. Here are "sar -u -M
5 5" and "sar -q -M 5 5" from one of the vpars.

sar -u -M 5 5

13:55:22 cpu %usr %sys %wio %idle

13:55:27 0 41 7 0 51

               1 45 6 1 49

          system 43 6 1 50

13:55:32 0 46 4 3 47

               1 37 13 1 49

          system 42 9 2 48

13:55:37 0 46 8 0 46

               1 53 13 0 34

          system 50 10 0 40

13:55:42 0 45 4 1 51

               1 39 12 0 49

          system 42 8 0 50

13:55:47 0 21 6 2 72

               1 30 5 1 64

          system 25 6 1 68

Average 0 40 6 1 53

Average 1 41 10 1 49

Average system 40 8 1 51

sar -q -M 5 5

14:44:50 cpu runq-sz %runocc swpq-sz %swpocc

14:44:55 0 8.8 100

               1 0.0 0

          system 8.8 50 0.0 0

14:45:00 0 4.0 40

               1 3.2 80

          system 3.5 60 0.0 0

14:45:05 0 31.0 20

               1 5.7 60

          system 12.0 40 0.0 0

14:45:10 0 8.0 100

               1 3.5 40

          system 6.7 70 0.0 0

14:45:15 0 3.7 60

               1 4.6 100

          system 4.2 80 0.0 0

Average 0 8.4 64

Average 1 4.3 56

Average system 6.5 60 0.0 0

This is also see in Top and Glance.

Glance

************************************************************************
********

                                CPU BY PROCESSOR
Users= 4

CPU State Util LoadAvg(1/5/15 min) CSwitch Last Pid

------------------------------------------------------------------------
--------

  0 Enable 60.8 18.4/ 18.2/ 18.8 2592 13070

1 Enable 60.8 3.0/ 2.8/ 3.1 3348 12726

Top

************************************************************************
********

Load averages: 10.35, 10.46, 10.90

429 processes: 389 sleeping, 39 running, 1 zombie

Cpu states:

CPU LOAD USER NICE SYS IDLE BLOCK SWAIT INTR SSYS

 0 17.75 31.7% 0.0% 0.0% 68.3% 0.0% 0.0% 0.0% 0.0%

 1 2.96 47.5% 0.0% 0.0% 52.5% 0.0% 0.0% 0.0% 0.0%

--- ---- ----- ----- ----- ----- ----- ----- ----- -----

avg 10.35 39.6% 0.0% 0.0% 60.4% 0.0% 0.0% 0.0% 0.0%

This is really uncharacteristic for any Unix server I have administered.
I see the CPU utilization about even most of the time, but the CPU 0 has
a very high run queue. No users are complaining and it seems that all
of the processes are getting run in a reasonable amount of time. Has
anyone seen this type of behavior before? Is it related to running in a
VPAR?

   Thanks

   -Jim

--
             ---> Please post QUESTIONS and SUMMARIES only!! <---
        To subscribe/unsubscribe to this list, contact majordomo@dutchworks.nl
       Name: hpux-admin@dutchworks.nl     Owner: owner-hpux-admin@dutchworks.nl
 
 Archives:  ftp.dutchworks.nl:/pub/digests/hpux-admin       (FTP, browse only)
            http://www.dutchworks.nl/htbin/hpsysadmin   (Web, browse & search)

• Previous message: Jeff Lightner: "[HPADM] rw as part of hostname for NFS export"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]