The first thing I notice is that you're not comparing the same data. A 14 minute interval and a 2 minute interval will provide dramatically different results particularly with something as bursty as a CPU.
-Kevin Casimir Couvillion wrote: > I am currently evaluating Servers Alive and trying to duplicate some > of the tools that I currently use. > > Presently, I use a Performance Monitor set to an 14 minute interval to > track processor utilization so I can look at 24 hour trends. I set > Servers Alive up to monitor the same WMI counters, but on a two minute > interval, then used RRDTool to log and plot the results. The output > graphs are markedly different between Servers Alive and Perfmon. I > have created a quick webpage <http://www.cpcouvillion.com/> with > examples of the graphs, note especially the SVR-MAIL lines. > > The Servers Alive/RRDTool output is very erratic in nature, whereas > the perfmon output is smoother with only the occasional spike during > expected high usage times like backups. If I look at the actual > results, two of the four machines often return a value of 0 for > Processor Usage for one or both. Performance Monitor doesn't seem to > record zero values, if I look at the counter, there is always a > minimum value greater than zero. > > Roundtrip times are on the order of 200 ms whether a zero is returned > or a value, timeout is left at default. I've got logging turned on, > but don't see any errors. All counters return non-zero values at > times, so it looks like they are set up at least partially correct. > The other counters will occasionally return a zero and all will > occasionally also return very small numbers like 0.00128. Servers > alive is being run as a domain admin, so permissions should be okay. > > From reading the Microsoft documentation > <http://msdn.microsoft.com/library/default.asp?url=/library/en-us/wmisdk/wmi/win32_perfrawdata_perfproc_process.asp>, > > Processor Usage is calculated over a "sample interval". It is unclear > to me if the interval is preset or part of the WMI query. Perhaps > perfmon is looking at an average over a large number of seconds and > Servers Alive is getting closer to instantanious results. > > I can smooth the data by plotting a trend, but I would like to know > why the data is so different. Are there tweaks or other methods to > get more consistent output? Am I trying to fix something that isn't > broken? > > Casimir > > > > To unsubscribe send a message with UNSUBSCRIBE as subject to > [email protected] > If you use auto-responders (like out-of-the-office messages), then > make sure that they are not send to the list nor to the individual > members of the list that send a message. Doing this will get you > removed from the list. To unsubscribe send a message with UNSUBSCRIBE as subject to [email protected] If you use auto-responders (like out-of-the-office messages), then make sure that they are not send to the list nor to the individual members of the list that send a message. Doing this will get you removed from the list.
