On Fri, 2004-03-12 at 16:51, Daniel Jarboe wrote: > So if the VM records show that a guest used 10% of the total CPU, and a > before/after of /proc/stat shows that half of the ticks were spent in > the "idle task", did the guest truly only need 5%? As more guests use > CPU, I'd expect that guest's usage to drop below 10% from z/VM's > perspective, and the "idle jiffies" to drop from linux's perspective, > but the user/nice/system numbers would not change much assuming similar > work was being done, and we aren't always running at 100%. Does that > sound about right?
No, I think it is a bit different. You're right in that the CPU usage reported by Linux will be reasonable correct as long as you have excess CPU capacity on VM (your monitor would report the % CPU wait). Separate from this aspect, the user/system ratio in Linux on zSeries is known to be too low when I/O is done. The reason for this is in architecture differences between zSeries and Intel. With Linux, idle is not really idle. But you should be able to get a bottom line measurement (e.g. 0.01% of a CPU). When you have plenty of CPU on VM and the guest does not run the CPU flat out 100%, it must be waiting for something (probably I/O, or maybe page faults on z/VM). When you start enough guests on VM to create contention it will cause the CPU to be taken away in the middle of the work, so the percentage of wall clock time where you have work that wants to run will increase. Rob ---------------------------------------------------------------------- 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
