Subject: Re: Poor timekeeping on x86 smp branch?
To: Sean Doran <>
From: Rafal Boni <>
List: tech-smp
Date: 04/23/2002 14:45:22
In message <>, you write: 

-> | 	I figured I'd send a heads-up if this wasn't a known problem; in
-> | 	case it *isn't* I can do some poking around, but unfortunately
-> | 	have too many other things on my plate to do any thorough digging.
-> How horrible is horrible?

*Really* horrible; lost 45 minutes over the course of 9 hours and ntpd
seemed to be primarily responsible for that badness... The clock in that
box is nowhere near as bad by itself.

-> I run a 386 SMP kernel from the same period, and I see a mean
-> hourly offset of |0.008| sec.  For an aging peecee in an imperfect
-> environment, this doesn't strike me as horribly bad time disciplining.
-> Do you see a difference between an SMP kernel's behaviour and
-> that of a single-processor kernel on that machine?

This is actually the first time I've seen this with the SMP kernel, 
though I don't recall if that box has been running ntpd before this
or not.  I hadn't seen any bad time behaviour before going to the
SMP kernel.


Rafal Boni                                           
  We are all worms.  But I do believe I am a glowworm.  -- Winston Churchill