On Thu, Oct 2, 2008 at 12:33 AM, Schuh, Richard <[EMAIL PROTECTED]> wrote:
> We have a situation where we need to discover what hot spots exist in > specific virtual machines, where they are spending the most execution time, > from outside the machines in question. Are there any tools that can do this > type of monitoring? Turning on traces in the machines has been rejected > because of the impact to the machines. The ideal is to have some kind of high-frequency sampling. With Linux, we enable the 100 Hz timer and record the old PSW on timer ticks. The overhead of measuring that is normally acceptable. Don't know whether TPF has such a thing (hope not, since it hurts performance). I've also done things in the past by analysis of the CP trace table. You need to cut down to trace just the runuser type entries for the user you profile, otherwise it gets too big. The sampling is not uniform, but since I think TPF does a lot of I/O, it may be frequent enough. -Rob
