RE: CPU count vs. CPU clock speed
- From: "Kevin Closson" <kevinc@xxxxxxxxxxxxx>
- To: <tanel.poder.003@xxxxxxx>, "ORACLE-L" <oracle-l@xxxxxxxxxxxxx>
- Date: Fri, 20 Jan 2006 12:11:52 -0800
>>>And the CPU time spent servicing interrupts (or at least
>>>their top halves) was reported under the unlucky process'es
>>>system time?
depends on the OS...if you consider Linux an OS, it
does "charge" the interrupted process with the interrupt
context execution time... it is a 2 edged sword. If you
only charge a process for the kernel mode cycles incurred
by whatever executes under it via syscall, who are you going
to charge? The process was on a CPU that spent a bunch of
time in kernel mode... now the problem is that it makes
the curious amongst us go off and look at the system calls
the process is making and we naturally impune those system
calls as "costing too much" for the process...when in fact,
there are cycles that had nothing to do with syscall...
--
//www.freelists.org/webpage/oracle-l
Other related posts:
