[2329] in SIPB_Linux_Development
Re: [Brandon S. Allbery KF8NH: Re: NTP dumps Linux, film at 11. [Fwd/FYI]]
daemon@ATHENA.MIT.EDU (Theodore Y. Ts'o)
Wed Dec 2 01:23:13 1998
Date: Wed, 2 Dec 1998 01:22:49 -0500
From: "Theodore Y. Ts'o" <tytso@MIT.EDU>
To: amu@MIT.EDU
Cc: "Theodore Y. Ts'o" <tytso@MIT.EDU>, linux-dev@MIT.EDU
In-Reply-To: Aaron M. Ucko's message of 02 Dec 1998 00:04:27 -0500,
<udlu2zfi17o.fsf@tux.mit.edu>
From: amu@MIT.EDU (Aaron M. Ucko)
Date: 02 Dec 1998 00:04:27 -0500
That's the first I've heard of clock skew nearly that bad.
I can replicate it easily using a BusToaster PCMCIA SCSI controller, and
running mke2fs -c on a JAZ disks. Apparently the Adaptec 1522
controller keeps interrupts disabled long enough that in the time to run
the mke2fs command, the clock will slow by over 5 minutes (guess how I
noticed).
But Brandon claims in his cluster of Linux machines at CMU that half the
machines were running fast, and half were running slow, and so the
explanation of losing clock interrupts doesn't fly.
The only thing I could think that made any sense was some kind of Linux
AFS time sync bug, but I would have thought other people (especially
here at MIT!) would have seen it as well.
- Ted
