The two powersaving technologies for CPUs called Cool'n'Quiet (AMD) or Speedstep (Intel) can cause monitoring products like our PRTG and IPCheck to take incorrect measurements, especially when PRTG is run on a virtual machine on such a hardware.
When we ran PRTG 6 on a virtualized Windows XP machine running on VMWare server on an Ubuntu/Linux host we saw strange peaks of traffic and gaps in the graphs. After some investigation we found that the internal timers of the Windows machine sometimes slowed down (compared to real time) and ran faster later to catch up. Thus the PRTG 6 monitoring station showed readings for the bandwidth measurements that were too high and too low (the counter deltas are divided by time to calculate the speed). In order to run PRTG 6 (and other products that rely on exact timing) on such a virtualized machine it is necessary to
Generally it seems to be a good idea to disable the two technologies when using monitoring system, regardless of virtual or real system.