FAQ3: How can I Monitoring Shared Processor Pools longer term?
nagger 100000MRSJ Comments (12) Visits (11914)
I had better point out that any performance tool of AXI command in a POWER Virtual Machine (LPAR) can only see its use of the Physical CPU, the size of its own CPU pool and (if Monitoring is switched on at the HMC), it has a number of the CPU use by "Other LPARs" use of its own CPU Pool. nmon gets these numbers by default and the nmon analyser graphs this data in the LPAR tab - this assumes its a Shared CPU LPAR. If it is a dedicated CPU LPAR then it can't get any of this Pool data and there is no LPAR tab in the nmon analyser spreadsheet (I get asked this all the time!).
But it you have anything other than the straight forward single Default CPU Pool then this limited picture does not show you want the rest of the machine is doing. This is primarily a security issue (data transmission "side channel") but its also complicated and CPU time consuming for every LPAR to know what all 64 CPU pools it doing - when its really none of its business!
The CPU Pool information is gathered (if switched on) at the HMC via the service processor but extracting it and then using it to generate graphs or spreadsheet data is extremely complex particularly if you have a complex set of up to 64 CPU Pools!
For a starter pack:
Take a look at the Redbook "IBM PowerVM Virtualization Managing and Monitoring" and search for lslparutil command (section 5.2.1 and page 104 in the current release) - it has an example of getting memory stats (for CPU Pools just change the field names you want) but it is a long coding path from there to having grap
hscroot@hmc11:~> lslparutil -m 8204-E8A-SN105B0C0 -r procpool -h 1 | tail -1
You will of course be wanting to remotely execute the comment to get the data off the HMC. Note: I added newlines to make it readable. You then need to do this once every 5 minutes and take the differences and divide time_cycles to get physical CPU numbers that you could start graphing. You would also need to store the numbers, reformat them and use a tool to graph them or create CSV files for a spreadsheet.
There is a further worked example on the lslparutil command manual page where (assuming you have the data collection switched on), you can get history information here shown hourly but it can go down to every 5 minutes or so.
lslparutil -m sys1 -r pool --startyear 2006 --startmonth 2
At the bottom is the calculations you need to make so the numbers make sense. The calculation is only for the first row of data. If you have multiple pools then you have to repeat this for each of the pools!!
There are plenty of other 3rd part tools too and, of course, ITM which can aggregate data - check that have HMC extraction agents because this is the only place you can get the data.
I hope this helps you either:
Best of luck, Nigel Griffiths