On Fri, 24 Jan 1997, Paul Goyette wrote:

> On Fri, 24 Jan 1997, Christoph Ewering wrote:
> 
> > Hello guys!
> > 
> > Could someone explain to me why it is so difficult for a Unix-system to
> > hold the time.
> 
> Well, on the Mac, the clock's interrupt isn't very high, relative to other
> things in the system (like SCSI and serial ports).  So, if the clock ticks
> while we're busy doing something else, we miss the interrupt.

Paul's explanation is correct.  (In fact, the clock interrupt interrupts
the system through VIA1, which is at IPL1 -- the lowest interrupt priority
on the system.)

> > I think Scott reported that his SE/30 looses 9 hours during
> > compiling a whole night. That's a lot.
> 
> Yep, but I'm pretty sure Scott is using the sbc SCSI driver, which AFAIK
> does not do interrupt-driven transfers (ie, Pseudo-DMA).  So, it spends
> large amounts of time with the clock interrupt blocked.

Well, it doesn't do interrupt-driven transfers, but it's become much less
of a problem with the new soft interrupt code I committed.  Note that I
said I previously lost about 90 minutes over 9 hours of heavy compile time
-- about 10 minutes per hour -- with the old code, and only about 2
minutes per hour with the new code.  I have not done any rigorous testing,
however, so your mileage may vary. :-)

--scott