Comments (2)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 localhost commented Permalink

Martin,while you're at it - one thing that has sat on my mind for a long time: CPENABLE. My proposition is hre that the FEWER CPs in an LPAR handle I/O interrupts, the better the OTHERS can pipeline - increasing processor speed, or #instructions/second.<div>&nbsp;</div> Based on many RMF I CPU reports I think the dfault value for z900/z9/z10 should have a much higher lower/upper number, i.e. you should try and NOT switch on a next CP to handle I/O interrupts until you have a SIGNIFICANT percentage of I/O interrupts queued (%TPI), and try and switch the CPs away from handlingI/O interrupts AS SOON AS POSSIBLE. Now this shuld always be a site-specific pair of values, but I'd like to see the difference between the CPENABLE default (10,30) and something that my intuition tells me should have animpact, like (60,80)....<div>&nbsp;</div> Would this ba of interest to you?<div>&nbsp;</div> regards<div>&nbsp;</div> Michael Weintraub,<div>&nbsp;</div> (If you meet Pat Ryan while @ITSO, give him my regards, please. Same for Mike Connoll.... I was there once, co-writing a Websphere monitoring etc. redbook...)

2 localhost commented Permalink

Michael, thanks for reading and commenting...<div>&nbsp;</div> I think I'll leave CPENABLE to the WSC folks as they manage to produce some advice on this about once a year. :-) Seriously this residency is focused on other things - Parallel Sysplex performance.