Originally posted by: cbartlett

Hey Guys,

I did a couple of searches but was not able to find this particular issue I am seeing on two of my LPARs.

I have a P750 (24 CPU 512G) AL730_049.
Dual VIO ioslevels: 2.2.0.10-FP-24
20+ LPARs running AIX 5.3, 6.1 and 7.1

The problem LPAR is AIX 6100-05-02-1034
CPU
Min 0.20
Entitled 0.40
Max 8.00
Virtual online 4
Logical online 16 (smt 4)

CPU shows 3.99 Physical CPU used
lparstat shows 66.4% idle.

Why is this LPAR not ceding idle cpu back to the shared pool? I cannot figure out how to determine what is not releasing the idle cycles.

root@stlpgareora1:/ # lparstat

System configuration: type=Shared mode=Uncapped smt=4 lcpu=16 mem=40704MB psize=20 ent=0.40

%user %sys %wait %idle physc %entc lbusy app vcsw phint

---

-----

---

------

---

-----

---

---

---

-----
24.9 5.1 3.6 66.4 3.94 985.3 16.7 6.65 6363412682 1124482820
Any help is appreciated. :)

-Caleb
Sr. AIX Systems Administrator
Reinsurance Group of America
#AIX-Forum

Originally posted by: The_Doctor

Please take the time to learn to use the code tags.... most that follow this forum don't have the spare cycles to decipher poorly formatted posts. So do yourself a favor:

• learn to use the code tages, and
• use the PREVIEW tab

BEFORE you POST your message. An easy to read post is more likely to generate better responses.

A snipite of what you wrote looks like:

root@stlpgareora1:/ # lparstat   System configuration: type=Shared mode=Uncapped smt=4 lcpu=16 mem=40704MB psize=20 ent=0.40   %user   %sys   %wait   %idle   physc   %entc   lbusy   app    vcsw       phint   24.9    5.1    3.6     66.4    3.94    985.3   16.7    6.65   6363412682 1124482820

In a virtual world, with SHARED processors, the %idle = 66.4% is pretty much a meaningless number. %user, %sys, %wait are also pretty much meaningless in a SHARED processor enviroment. You are probably best served if you just ignore this numbers, as most tend to get confused by them.

The number that is meaningful is physc = 3.94 which tells me your 4 VPs are pretty much at 100%, therefore there are 0 (or next to zero) cycles to cede back to the shared processor pool.
#AIX-Forum

Originally posted by: cbartlett

Hey Doc,

You'll notice that was my first post, so you'll have to forgive the formatting. If it were my 80th post I could understand your aggravation. I did attach a nmon screenshot. In the screenshot you will see the CPU's are not that busy at all (topas, ps, etc.. indicate the same) and are mostly idle. On each of my 50 LPAR's spread across two P750's I have only two LPARs exhibiting this behavior. A lot of the CPU is idle and not ceded back to the shared pool. These are oracle systems so I've disable oracle startups, removed all performance collecting and system monitoring so I have only the OS running after reboot and still see 4 physical CPU's being consumed. Obviously not right. I don't agree that the idle time is meaningless. It should be ceded back to the pool. It is not. I am concerned those cycles may not be available to the other LPARs. Galileo and Ganglia show the idle CPU as in use so even if the number is meaningless, it renders my performance data useless. When budgeting for oracle licenses and sizing systems I need this data to be accurate and uniform especially if we start to charge a cost center for CPU consumption.

Attached you will find a Galileo screenshot. The Red and Blue are the two LPARs showing a lot of idle time. It appears those idle cycles are not available to the rest of the pool. This appears to consume a lot of physical CPU on the frame. Performance on these guys is awesome even during month end when utilization is much higher than it is right now. Either way, I need to show that idle CPU available to the pool.

And I'll learn the code if I plan to post further, Thanks for your time! -Caleb
#AIX-Forum

Originally posted by: cbartlett

I've also changed the VP to 8 and the LPAR running only a OS will take 8 physical CPU again... very little user and system utilization with a huge amount of idle.
#AIX-Forum

Originally posted by: The_Doctor

so from your new description of the problem, can we assume that the "wait" processes are responsible for the excessive CPU ticks ? If so, then you probably have the problem discussed here:

• Link: http://www.ibm.com/developerworks/forums/thread.jspa?messageID=14690189

and

• Link: http://www-01.ibm.com/support/docview.wss?crawler=1&uid=isg1IV01111

You can also google -> WAITPROC IDLE CONSUMES CPU AIX <- if the direct link to the APAR fails to work for you.

But don't rely on my WAG, you'll need to come to your own conclusion(s) how close this matches your own circumstances.
#AIX-Forum

Originally posted by: cbartlett

Thanks Doc! I'll see if these help lead me to a resolution and will let you know. Thanks again for your time. :)
#AIX-Forum