Re: Reported CPU utilization of network server doing constant work - varies

[Mike Spreitzer]

Mystery solved due to a helpful response from lkml member Mark Hahn, who 
cited the McCanne & Torek paper in the USENIX Winter Conference '93.  My 
clients were synched with their 10 ms clock, which was beating against the 
10 ms clock used on the IA32 server to sample CPU utilization.

McCanne & Torek discuss a better CPU accouting mechanism they put in BSD. 
AIX also apparently does something better (or the RS/6000 clock is much 
worse).  Hint, hint.

Thanks,
Mike