prmmonitor

[ITSD-ACCS]

What is the meaning of prmmonitor column CPU used ?

CPU CPU PRM Group used
---------------------------------------------
OTHERS 1 10% 0.0%
admin 2 30% 0.5%
outdoor 3 35% 0.0%
indoor 4 25% 5.6%

[Michael Steele_2]

These are thresholds. PRM CPU Resource Manger Ensures that each PRM group is granted at least its share, but no more than its capped amount of CPU.

You can change these with "xprm".

Note: I've heard of a PRM bug where, when used with ServiceGuard as a failover mechanism the Resources have to max out at 100% first, failover, and only then will the thresholds get put into effect.

[ITSD-ACCS]

Thanks for your reply.

CPU CPU
PRM group PRMID Entitlement Used
---------------------------------------------
OTHERS 1 10% 0.0%
admin 2 30% 0.5%
outdoor 3 35% 0.0%
indoor 4 25% 5.6%

For the above sample , what is the meaning of 0.5 % and 5.6% ? Is it the loading % of overall CPU ?

In case the system at that time is 55% loading ( using sar with 45% idle ). Does it mean that the outdoor PRM group has consume 35% of 55% loading ?

[Michael Steele_2]

You have four groups

1 - OTHERS
2 - admin
3 - outdoor
4 - indoor

OTHERS is pre-allocated only 10% of the overall CPU resources. For example, if you have 10 CPU's then OTHERS has one.

admin is pre-allocated only 30%, outdoor 35% and indoor 25%.

10% + 30% + 35% + 25% = 100%

For OTHERS, out of the 10% pre-allocated it is using 0.0%. For admin 0.5%, outdoor 0.0%, indoor 5.6%.

sar -u 5 5 is the CPU metric of choice used by HP RCE's.

A CPU bottleneck is a combination of 3 factors: 100% utilization, 1 or more process in the run queue and a large number of processes running. Use sar -u 5 5 to measure your process table. Use vmstat and note the number of jobs in the run queue.

Like measureware, PRM is a resource consumer that doesn't always justify its running under a heavily loaded server. For example, stopping measureware usually returns 5% back to the CPU. I'm sure there's a similar number for PRM.

See:

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B8733-88007/B8733-90007_top.html&con=/hpux/onlinedocs/B8733-90007/00/00/10-con.html&toc=/hpux/onlinedocs/B8733-90007/00/00/10-toc.html&searchterms=prmmonitor%7cxprm&queryid=20030307-113801

[Jonathan Fears]

The 'CPU Used' column tells the percentage of CPU consumed by each group during the specified interval. A group's resource use can be less that its entitlement when there are not enough ready processes in that group requesting the resource. The above output gives details about all groups except PRM_SYS. PRM internally creates this group with PRMID 0, and assigns system processes to it. Root logins are also placed in this group unless a user records indicates otherwise.

To see prmmonitor output with the PRM_SYS group listed, use the command "prmmonitor -s". This starts prmmonitor output at PRMID 0 instead of PRMID 1 (OTHERS).

[Dan Herington]

Hi,

A couple of clarifications.

The "used" number is the amount of CPU that was actually consumed by the processes running in each group. The Entitlement is how much it is guaranteed, if it needs it.

In response to the previous message, PRM takes very little CPU itself. Our design goals are to keep this below 3%, but we have never seen it get that high in our tests.

In reading these messages, it appears that you have a system with 55% utilization, but prmmonitor is only showing 6%. There is a special group called PRM_SYS that is used for root processes and OS daemons. This group gets as much as 50% of the system, but rarely ever uses more than a few %. You can get prmmonitor to show this group in its output by including the -s option.

My guess is you have some process that was started by root that is running out of control.

Dan