From: Rick DeNatale <rick.denatale@...>
Date: 2006-10-27T08:02:56+09:00
Subject: Re: simple math question

On 10/26/06, M. Edward (Ed) Borasky <znmeb@cesmail.net> wrote:
> Paul Lutus wrote:
> > Rick DeNatale wrote:
> >
> > / ...
> >
> >> By the way, is the actual meaning of user, system, total, and real
> >> documented anywhere.  I've looked around a few times, and haven't
> >> uncovered it.  I guess I could try to devine it by reading the code
> >> for benchmark, but...
> >
> > System time is time spent outside userspace, e.g. direct kernel calls.
> > User time is time spent directly executing user code in userspace.
> >
> > I think total time and real time are self-explanatory, but just in case ...
> > total time is system + user + kernel overhead not properly accounted for,
> > and real time is simply an RTC measurement at the end of the process life
> > subtracted from the same at the beginning. It is expected (in a manner of
> > speaking) that none of these quantities will add as one would expect.
> >
>
> I actually dug into this a few weeks ago for the RHEL 3 version of the
> 2.4.21 Linux kernel. It's mildly complicated and changes from version to
> version of the Linux kernel, especially the scheduler. And the external
> tools don't always "do the right thing" either. And I *only* looked at
> Linux; I'm not in possession of the same information for BSD or MacOS. I
> can probably track it down for Windows, however -- it's documented in
> Mark Friedman's book in the Windows 2003 Server Resource Kit.
>
> M. Edward "That Guy" Borasky
>
> P.S.: Paul's response is essentially correct and "close enough for all
> practical purposes". I get paid to care about the cases when it isn't. :)
>

Thanks guys, I figured that it was something like that. I looked at
the source code for benchmark and it's just getting Process.times
before and after the block.

That said, I wonder what these presumably purely computational blocks
are doing in the kernel at all.  In a way that's why I wasn't sure, in
the context of ruby benchmarking I was sort of hoping that it was
differentiating between library C code and ruby code, but it really
isn't.

I guess that the system time must be spent in something like process
switching overhead, yes?

-- 
Rick DeNatale

My blog on Ruby
http://talklikeaduck.denhaven2.com/