Re: personal webpage for L4-Hurd

[OKUJI Yoshinori]

From: address@hidden (Niels Möller)
Subject: Re: personal webpage for L4-Hurd
Date: 23 Nov 2000 18:30:32 +0100

> What should "system time" mean under the hurd? I mean, the values
> reported to ordinary processes using rusage and similar functions. A
> reasonable (although perhaps hard to implement) definition is that
> system time is the cpu time spent by Hurd servers on the process'
> behalf (where "Hurd servers" is some subset of all processes).

  That sounds reasonable, but it is still vague what should be
accounted as the system run time of a thread. For example, I guess
that most of UNIX-like operating systems don't assume that syslogd is
part of the systems. On Hurd, syslogd can be implemented as a
translator and as a daemon, so the definition of system run time
changes, depending on the implementation of syslogd, if you think
translators are part of the system.

> Perhaps
> the time spent inside the micro-kernel itself is of secondary
> importance? I imagine the cpu time spent inside the micro-kernel will
> be significantly smaller for L4 than Mach.

  That should be right, but I think it is desirable to have a way to
get kernel-level run time, when profiling threads which use system
calls intensively.

Okuji