Re: Giving more than 110%

[Larry Bolhuis <lbolhuis@xxxxxxxxxx>]

Dave,

This goes against what I thought I understood. Of course I could easily 
be wrong so I just want to be clear.

It has always been my understanding that if I had 1 virtual processor 
and 0.10 processor assigned to a partition I would still see the 
processor utilization on a 0 to 100% scale even though I techincally can 
only use 10%. In WRKSYSACT I would see the process QHLPVTASK using 9 
times the amount of CPU that the rest of the system was using. So if my 
actual processor utilization was 0.05 then I would see QHLPVTASK using 
45% of a processor so that WRKACTJOB/WRKSYSSTS shows 50%. (After all in 
this case I'm using 50% of what I've assigned to the partition)

The other rules as you explain them are exactly as I understood so I'd 
like to know if there is something I am misunderstanding.

Thanks,

- Larry

Dave Schnee wrote:
> Kirk:
>
> What probably happened to you was that you were on a System i5 machine 
> with "uncapped processors" and LOTS of unused CPU power in the processor 
> pool.  In that instance, you can "borrow" CPU cycles from other 
> partitions' allocations if you have your "virtual processors" number set 
> higher than the actual processor(s) you are assigned.  Since you must have 
> at least 0.10 processor assigned for each virtual processor you assign, 
> the most you can get is 1,000% (even if there are more processors still 
> idle) because i5/OS will only dispatch as many concurrent jobs as there 
> are virtual processors.
>
> In your example, you were assigned 0.50 processing units and could, 
> therefore, assign up to 5 virtual processors (simultaneous threads 
> executing).  That would allow you to use 1,000% of the CPU you were 
> assigned.  If there were not enough idle CPU cycles in the processor pool, 
> you would have used less than 1,000% despite your assignment of virtual 
> processors.
>
> Had you assigned 1 virtual processor with 0.50 processor units, you could 
> have gotten only to 200% CPU usage if there were otherwise-idle CPU cycles 
> (due to other partitions not using their full allocations).
>
> There is a Red(something) - either a RedPaper of RedBook - I'm not sure 
> which it will become - that is currently being reviewed before publication 
> by IBM.  I am one of the authors and there is a rather complete treatise 
> on this topic in that writing.  The publication number has been assigned 
> as SG24-7174 and the current working title is "Automating Dynamic LPAR on 
> iSeries and System i5 utilizing APRM".  I hope it gets "out" soon - this 
> is not a well-known concept.
>
> Dave Schnee,
> Barsa Consulting Group, LLC
>