There is a GSE conference for Linux, z/VM and z/VSE in Stuttgart Germany in 3 weeks. That same week, there is a performance class that will show how to get exactly what you want using zVPS "http://velocitysoftware.com/zvps.html". If you care about performance of Linux under z/VM, you need the z/VM data, the LPAR data, the virtual machine data, and especially the linux data down to the process level. zVPS collects, processes and displays all of this data.
The product can be used to analyze our data at "http://demo.velocitysoftware.com/ZVIEW/zview.cgi". click on the "linux reports", then on "esalnxp" to show all the linux processes active across our LPAR. close the menu then for better viewing. On 9/30/2015 1:45 AM, Mikael Wargh wrote:
Hello, This is probably an old already well digested topic, but is there any reliable way to monitor CPU usage on Linux level? Linux tools like TOP and SAR seems to show z/VM's cap as a maximum in their statistics eg. share is set to 4% per IFL --> 2xIFL is 8%. This is ok from providers point of view (us), but when generating graphs and lists to the customers they are wondering why they never get 100%. Also hardware changes etc. changes the percentage so it would be better not to show it to the end customer to avoid unnecessary confusion. Nmon seemed to show the capacity in a different way, at first it looked like that nmon somehow would understand this as graphs showed spikes up to 100%, but with further investigation it seems to be somehow screwed up when having multiprocessor environment. When stressing one Linux thread nmon shows utilization to be near the z/VM's cap, but when having second thread (two logical processors defined to this Linux) it jumps to 100%... When having uncapped Linux the nmon show CPU consumption as it should: one thread is 50% and two 100%. Is there any simple way to get the information or correctly scale the information so that we would get reports to the customer showing in 0-100% range? Customers naturally like to optimize their capacity and now it's a bit hard as they get some z/VM cap percentages as total usage statistics and they don't understand the physical layer beneath. Maybe one way to optimize would simply to be by focusing and minimizing the steal percentage shown in TOP or SAR, but the CPU statistics would still be uncorrected. Br, Mikael Wargh Tieto Oyj
---------------------------------------------------------------------- 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/