clock steps backwards (was Re: svn commit: rev 2845 - trunk/subversion/tests/libsvn_fs)

On 2 Aug 2002, Philip Martin <philip@codematters.co.uk> wrote:
> Mix <mixtim@acm.org> writes:
>
> > > $ ./a.out
> > > 43043: 374436 < 374437
> > > 524931: 904361 < 904362
> > > 778682: 184323 < 184324
> > > 887643: 304306 < 304307
> > > 1014906: 444288 < 444289
> > > 1187703: 634262 < 634263
> > > 1296827: 754245 < 754246
> > > 1315004: 774241 < 774242
> > > 1442331: 914224 < 914225
> > > 1596600: 84201 < 84202
> > > so yes it goes backwards.
> >
> >
> > I hate to tell you guys this, but you have screwy hardware or an OS
> > bug here as this isn't possible. I use gettimeofday() a bit at work
> > for some performance testing and this has never happened to me.

One reason why this can happen is that on machines with variable clock
rates (e.g. Intel's SpeedStep, common on laptops), the kernel gets
confused about the relative speeds of the realtime clock and the CPU's
cycle counter. This happens on my T20 Thinkpad and it gives X11 fits.

You might like to try using the "notsc" boot option (on 2.4.18 at
least.) There are some other reasons why it might happen, and this
won't fix all of them.

-- 
Martin 
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org