I know I know - first round on you - because the chocolate install of zvm has that in the directory entry? Do I get the first round?
These Linux workloads especially the polling apps are a brave new world for the z/VM scheduler. Your papers lays it out there nicely. When you need to find the sweet spot of CPU it takes some finagling. David -------- Original Message -------- Subject: Re: Set Share Relative From: Scott Rohling <scott.rohl...@gmail.com> Date: Tue, June 14, 2011 6:51 pm To: LINUX-390@VM.MARIST.EDU No worries, Sir Rob - I seem to be especially cranky today. Excellent paper and explanation of relative share - and a much better answer then the simplistic HELP explanation. I'll refer to this in the future... Thank you! Scott Rohling On Tue, Jun 14, 2011 at 2:56 PM, Rob van der Heij < rvdh...@velocity-software.com> wrote: > On Tue, Jun 14, 2011 at 7:57 PM, Scott Rohling <scott.rohl...@gmail.com> > wrote: > > > My statements were based on the help for CP SET SHARE.. my comments on > math > > were somewhat tongue in cheek - but since this is a relative value - 100 > is > > as good as any other to base things on.. set all your guests to 7 and > use > > increment of 3 if you like. No idea at all by what you mean by > 'political' > > .. and annoyed at the statement that the defaults are just flat out > wrong. > > My apologies. One should not post on the mailing list while standing > up or while in a hurry... > > You can't do VM performance with just the CP help files. I provided > the link to the PDF that explains that it does not work the way you > might think it should. The reason you see relative share like 100 and > 200 is not for simple math or aesthetics, but to compensate for the > fact that the allocated share is distributed over the virtual CPUs. > The paper explains what the implications are when you don't play the > game right. > > We're pushing the scheduler to its limits with workloads that are > completely different from what it was designed for. It works as long > as you don't do silly things or try to mislead the scheduler. > Political tuning is when someone claims that VTAM is the most > important thing in the system and recommends it should have REL 10000 > (or the z/OS sysprog insist that their OSA port should be "preferred > route" even though he has no clue...) > > As for "wrong" - I thought that was beaten to death already. I buy an > adult beverage for the first* who can explain why it makes sense to > have the TCPIP stacks with a relative share 30 times higher than their > Linux production guests. > > Rob > -- > Rob van der Heij > Velocity Software > http://www.velocitysoftware.com/ > > ---------------------------------------------------------------------- > For LINUX-390 subscribe / signoff / archive access instructions, > send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or > visit > http://www.marist.edu/htbin/wlvindex?LINUX-390 > ---------------------------------------------------------------------- > For more information on Linux on System z, visit > http://wiki.linuxvm.org/ > ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/
