Measuring CPU performance? Which is right?
On the zLinux guest (ZP013), using sar I get a CPU usage of about 15%: 11:00:01 AM CPU %user %nice %system %iowait%steal %idle 11:50:01 AM all 14.78 0.00 0.70 0.53 0.1583.84 But under Perfkit (zVM) we get the following exception message, 33.5% CPU: 11:51:51 FCXUSL317A User ZP013 %CPU 33.5 exceeded threshold 30.0 for 5 min. 'top' numbers match the 'sar' values from the zLinux guest. We have two IFLs defined to the guest. Why are the numbers from PERFKIT different from the zLinux environment? Which numbers should we be monitoring from? James Chaplin Systems Programmer, MVS, zVM zLinux Base Technologies, Inc (703) 921-6220 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Measuring CPU performance? Which is right?
On Thu, Jun 12, 2008 at 12:31 PM, in message [EMAIL PROTECTED], CHAPLIN, JAMES (CTR) [EMAIL PROTECTED] wrote: -snip- Why are the numbers from PERFKIT different from the zLinux environment? Which numbers should we be monitoring from? You didn't say which version of which distribution you are using. Anything prior to SLES10 and RHEL5 will report (possibly wildly) incorrect CPU utilization numbers from inside the Linux guests. If you're at one of those levels or higher, then the numbers should match. Mark Post -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Measuring CPU performance? Which is right?
CHAPLIN, JAMES (CTR) writes: On the zLinux guest (ZP013), using sar I get a CPU usage of about 15%: [...] But under Perfkit (zVM) we get the following exception message, 33.5% CPU: 11:51:51 FCXUSL317A User ZP013 %CPU 33.5 exceeded threshold 30.0 for 5 min. [...] We have two IFLs defined to the guest. [...] Why are the numbers from PERFKIT different from the zLinux environment? PerfKit percentages are calculated as percentage of one engine. Linux percentages calculate percentage of CPU resource available to the image. For your Linux guest with 2 engines, Linux tells you it's using ~15% of its 2-engines'-worth. PerfKit spells that as ~30% of a nominally-100%-utilised single engine. Same resource usage, different way of displaying the measurement. [For the purposes of this posting, I'm treating any remaining few percent difference as a second order effect or else we'd muddy the waters with discussing a bunch of more complex measurement issues.] --Malcolm -- Malcolm Beattie System z SWG/STG, Europe IBM UK -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Measuring CPU performance? Which is right?
Malcolm Beattie wrote: CHAPLIN, JAMES (CTR) writes: On the zLinux guest (ZP013), using sar I get a CPU usage of about 15%: [...] But under Perfkit (zVM) we get the following exception message, 33.5% CPU: 11:51:51 FCXUSL317A User ZP013 %CPU 33.5 exceeded threshold 30.0 for 5 min. [...] We have two IFLs defined to the guest. [...] Why are the numbers from PERFKIT different from the zLinux environment? PerfKit percentages are calculated as percentage of one engine. Linux percentages calculate percentage of CPU resource available to the image. For your Linux guest with 2 engines, Linux tells you it's using ~15% of its 2-engines'-worth. PerfKit spells that as ~30% of a nominally-100%-utilised single engine. Same resource usage, different way of displaying the measurement. Isn't that in the wrong direction? I'd expect 15% of two engines ~= 7.5% of one engine (.15x200 = .075x100). If, however, Linux is reporting as if 1 engine and PerfKit is doubling due to two engines, I could see your scenario. Which is correct? TIA Kim -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 begin:vcard fn:Kim Goldenberg n:Goldenberg;Kim org:State of New Jersey;Office of Information Technology (OIT) adr:200 Riverview Plaza;;PO Box 212;Trenton;NJ;08625-0212;USA email;internet:[EMAIL PROTECTED] title:Systems Programmer I tel;work:609-777-3722 tel;fax:609-777-3939 x-mozilla-html:FALSE url:http://www.state.nj.us version:2.1 end:vcard
Re: Measuring CPU performance? Which is right?
On Thu, Jun 12, 2008 at 3:19 PM, in message [EMAIL PROTECTED], Kim Goldenberg [EMAIL PROTECTED] wrote: -snip- Isn't that in the wrong direction? I'd expect 15% of two engines ~= 7.5% of one engine (.15x200 = .075x100). If, however, Linux is reporting as if 1 engine and PerfKit is doubling due to two engines, I could see your scenario. Which is correct? If using top, it varies depending on whether you have it show you all the CPUs available, or take the summary. If it is showing the summary line, it shows the average of the various CPUs. If it is showing the individual CPUs, then it lists how busy each one is. Some distributions also have the total/average of the individual CPUs listed as a separate line on the display. Mark Post -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Measuring CPU performance? Which is right?
RHEL 5.0 James Chaplin Systems Programmer, MVS, zVM zLinux Base Technologies, Inc (703) 921-6220 -Original Message- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Mark Post Sent: Thursday, June 12, 2008 1:33 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Measuring CPU performance? Which is right? On Thu, Jun 12, 2008 at 12:31 PM, in message [EMAIL PROTECTED], CHAPLIN, JAMES (CTR) [EMAIL PROTECTED] wrote: -snip- Why are the numbers from PERFKIT different from the zLinux environment? Which numbers should we be monitoring from? You didn't say which version of which distribution you are using. Anything prior to SLES10 and RHEL5 will report (possibly wildly) incorrect CPU utilization numbers from inside the Linux guests. If you're at one of those levels or higher, then the numbers should match. Mark Post -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Measuring CPU performance? Which is right?
On Thu, Jun 12, 2008 at 7:29 PM, Malcolm Beattie [EMAIL PROTECTED] wrote: PerfKit percentages are calculated as percentage of one engine. Linux percentages calculate percentage of CPU resource available to the image. For your Linux guest with 2 engines, Linux tells you it's using ~15% of its 2-engines'-worth. PerfKit spells that as ~30% of a nominally-100%-utilised single engine. Same resource usage, different way of displaying the measurement. Obviously Linux on z/VM has no idea how much is really available to it. It's just dividing things by the number of virtual CPUs. That does not mean Linux would have n * 100% available. [For the purposes of this posting, I'm treating any remaining few percent difference as a second order effect or else we'd muddy the waters with discussing a bunch of more complex measurement issues.] You can ignore what you like, but the remainder is probably significant. With the later kernel releases that use the virtual cpu accounting (recognized by the reporting of steal percentage) the basis is virtual time rather than total time. So anything done by CP on behalf of this Linux virtual machine is reported by Linux as unused. Rob -- Rob van der Heij Velocity Software http://velocitysoftware.com/ -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
Re: Measuring CPU performance? Which is right?
From zLinux, I am seeing one CPU at 15%, the other at a much lower value (~5%) using 'mpstat -P ALL' James Chaplin Systems Programmer, MVS, zVM zLinux Base Technologies, Inc (703) 921-6220 -Original Message- From: Linux on 390 Port [mailto:[EMAIL PROTECTED] On Behalf Of Kim Goldenberg Sent: Thursday, June 12, 2008 3:19 PM To: LINUX-390@VM.MARIST.EDU Subject: Re: Measuring CPU performance? Which is right? Malcolm Beattie wrote: CHAPLIN, JAMES (CTR) writes: On the zLinux guest (ZP013), using sar I get a CPU usage of about 15%: [...] But under Perfkit (zVM) we get the following exception message, 33.5% CPU: 11:51:51 FCXUSL317A User ZP013 %CPU 33.5 exceeded threshold 30.0 for 5 min. [...] We have two IFLs defined to the guest. [...] Why are the numbers from PERFKIT different from the zLinux environment? PerfKit percentages are calculated as percentage of one engine. Linux percentages calculate percentage of CPU resource available to the image. For your Linux guest with 2 engines, Linux tells you it's using ~15% of its 2-engines'-worth. PerfKit spells that as ~30% of a nominally-100%-utilised single engine. Same resource usage, different way of displaying the measurement. Isn't that in the wrong direction? I'd expect 15% of two engines ~= 7.5% of one engine (.15x200 = .075x100). If, however, Linux is reporting as if 1 engine and PerfKit is doubling due to two engines, I could see your scenario. Which is correct? TIA Kim -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 -- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390