Hmm, this sounds a good way to do it too. We're already gathering some IND based reports but in much coarse intervals and without further reprocessing. Maybe this would be the easiest way to handle the CPU part of the reporting. Thanks Tore! :) Looking at the big picture I still think that the z/VM <--> Linux programming/monitoring interfaces should be simpler and better documented. Maybe the upcoming KVM will change this. After all zLinux is moving more and more to open source world and getting good application development also on this platform needs more openness and community support.
-Mikael -----Original Message----- From: Linux on 390 Port [mailto:LINUX-390@VM.MARIST.EDU] On Behalf Of Agblad Tore Sent: 1. lokakuuta 2015 11:58 To: LINUX-390@VM.MARIST.EDU Subject: Re: zLinux CPU monitoring Hi, yes you will need figures from z/VM. And do some work yourself as already mentioned. We use IND output for all zlinux servers from z/VM every 5 minute saved into a file. You get total cpu seconds used since boot. So we read that generated file every night, and calculates difference for each 5 minutes interval giving us promille cpu used (better than percent, we can stick to integer). We put that (and memory usage as well by the way, active that is, Proj. WSET) in a mysql database. This is actually a c-pgm to get descent performance with the calculating :-) Every Monday we run a 'batch' generating html-pages including graphs for last week per server. So we can take a look with almost zero response time. And we can produce statistics and graphs whatever you want. ________________________________________________ Tore Agblad zOpen, IT Services Volvo Group Headquarters Corporate Process & IT SE-405 08, Gothenburg Sweden E-mail: tore.agb...@volvo.com http://www.volvo.com/volvoit/global/en-gb/ -----Original Message----- From: Linux on 390 Port [mailto:LINUX-390@VM.MARIST.EDU] On Behalf Of Mikael Wargh Sent: den 1 oktober 2015 7:37 To: LINUX-390@VM.MARIST.EDU Subject: Re: zLinux CPU monitoring This was good information, many thanks Tomas! As we're going more open with LinuxOne, KVM and probably more to come, I really think that at least the basic monitoring should have several options to choose from. In traditional x86 Linux world the performance monitoring sector is flooded by different free choices. Of course the userbase is larger by several magnitudes, but at least the z hardware platform is more homogenous than the variety on distributed side. That makes me think that easily accessed and used monitoring stream interface shouldn't be too hard to integrate on z/VM and zLinuxes. Now it seems that there are one mentioned below, but by quickly reading the presentation, the interface don't seem to be very elegant and requires quite much of tuning. Anyway I'm very interested to test it. :) Maybe all this could be done in z/VM, but personally I'm not very fond of writing REXX execs there... -Mikael -----Original Message----- From: Linux on 390 Port [mailto:LINUX-390@VM.MARIST.EDU] On Behalf Of Pavelka, Tomas Sent: 30. syyskuuta 2015 14:49 To: LINUX-390@VM.MARIST.EDU Subject: Re: zLinux CPU monitoring I was once researching something similar (i.e. how to get reliable CPU readings from within Linux) but unfortunately never finished. But I still have the links, maybe you will find something interesting there: Presentation about how Linux kernel can get accurate CPU readings from the hypervisor: http://linuxvm.org/present/SHARE110/S9266ms.pdf How to set up the hypervisor file system mentioned in the presentation above: http://www-01.ibm.com/support/knowledgecenter/linuxonibm/com.ibm.linux.z.lgdd/lgdd_r_hypfs_setup.html s390 Debug fs (I think this is where hyptop reads from): https://www.kernel.org/doc/Documentation/s390/s390dbf.txt Also look at /proc/[pid]/stat in proc man page: http://linux.die.net/man/5/proc None of these will give you readings without additional work (like computing percentages from cumulative readings). HTH, Tomas
