Topic
  • 1 reply
  • Latest Post - ‏2013-10-15T10:54:25Z by nagger
RFS
RFS
29 Posts

Pinned topic Question about NMON LPAR Tab Graphs

‏2013-08-23T11:29:39Z |

I have the following FRAME/LPAR configuration:

FRAME: IBM 9117-MMC P7 770
CPU Type: PowerPC_POWER7
CPU Speed: 3304 MHz
 

AIX Version: 6.1.8.15
OS Level: 6100-08-02-1316
LPAR vCPUs: 10

Type: Dedicated-SMT-4
Mode: Donating
Memory: 98304 MB

I captured data using "/usr/bin/topas_nmon".

The "Physical vs Entitlmenet" graph indicates I am using 5 cpu units of 10 that are avaialble or 50%. However the "CPU% vs VPs" graph is only showing most peaks around 15%.

Which graph is correct?

See attached graphs.

Thanks in advance for your feedback. :-)

 

 

 

Updated on 2013-08-26T10:53:19Z at 2013-08-26T10:53:19Z by RFS
  • nagger
    nagger
    1679 Posts

    Re: Question about NMON LPAR Tab Graphs

    ‏2013-10-15T10:54:25Z  

    Hi,

    With only the two graphs and you using this fairly rare mode I  will have to make some guesses.

    For CPU Donating LPARs you do not really have Entitlement not Virtual Processors - you specify the number of CPU like Dedicated CPU mode.

    For sanity reasons they make the CPU count = Entitlement and I would expect the Physical CPU Use graph to be correct. BUT if you only have a few processes running they will only be using the first SMT thread of the cores being used. One could argue that if one SMT thread is uses then the core is used and that is what is being shown here. On the other hand the core may be 100% used but if not using SMT thread 2, 3 and 4 it could be seen as having lots of unused capacity. So you could say the core utilisation is both 100% and say at a guess 65% used.

    For about the millionth time let me state Utilisation numbers with SMT on are very confusing and even worse if you are using uncapped shared CPUs with VP higher than Entitlement. In this case I think you have SMT1=say 75%, SMT2=5% SMT3=0% and SMT4=0% - if these are averaged the Utilisation is ~20% even when we know the core is ~100% used.

    Only use Utilisation numbers to check the User time is much larger than system (kernel) time. Then use Physical CPU use/consumed for monitoring your LPAR.   Utilisation is dead, long live Physical Use !!

    I hope this helps, cheers Nigel Griffiths