Date: Tue, 23 Mar 2004 11:36:28 -0600 From: Dan Nelson <dnelson@allantgroup.com> To: Chris Landauer <cal@rushg.aero.org> Cc: freebsd-questions@freebsd.org Subject: Re: wraparound value for (user) time Message-ID: <20040323173627.GC2492@dan.emsphone.com> In-Reply-To: <200403231730.i2NHU0r09531@calamari.aero.org> References: <200403231730.i2NHU0r09531@calamari.aero.org>
next in thread | previous in thread | raw e-mail | index | archive | help
In the last episode (Mar 23), Chris Landauer said:
> you werote:
> > Which value wrapped? user, system, or elapsed?
>
> > The best I could come up with was that elapsed time might be stored
> > in a long variable in milliseconds, which would wrap at 49.7 days.
> > User and system times are stored as "struct timeval"s and should
> > never wrap.
>
> thanx for answering so quickly - it is the user cpu time that wraps,
> in much less than 49 days, apparently (elapsed time was 164 hrs and
> 340 hrs in the two cases that wrapped, and the cpu percentage should
> have been between 90 and 95)
>
> i did finally find struct timeval, which has a long - arithmetically,
> i convinced myself that the value had to be not an unsigned one (so
> the wrapped values were negative, but printed as positive), but i
> need to look further at the resolution issue - i will study it
> further and report the results as soon as i have something definitive
> (or at least useful) to say
Note that struct timeval has two members, tv_sec, and tv_usec. tv_sec
will wrap in 2038, and tv_usec wraps every second :) It might be that
there's a bug somewhere in tcsh's time calculations; see if
/usr/bin/time does any better. Another thing to try is enabling
process accounting ("accton /var/account/acct"), and then use "lastcomm
-esu" to print the stats.
--
Dan Nelson
dnelson@allantgroup.com
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040323173627.GC2492>