Michael G. Schabert wrote:
[ running w/o rebooting loses time... ]

> I am of the school that runs all computers 24*7*365, so I definitely have
> left older computers running for weeks, but I do reboot at least once a
> week, on principle. Your RAM table must have looked like swiss cheese
> running 3 weeks without a reboot!

probably so, but i had several important things i was in the middle of
that i didn't want to mess with ;-)

> I constantly install and play with alpha,
> beta and developer releases of CDeVs and extensions, which require me to
> reboot every now & then (for the install, not the crashes ;-). Also, I used
> to change ISP settings fairly often when using 2 ISPs for a couple years.

understandable.

> I'm assuming that the reason that you're asserting that it must be done
> without rebooting is that the RTC only re-read on boot.

more or less (i imagine it could be read at other times, but i know for
sure that it is read at reboot).

> That means that the
> RTC is keeping the correct time regardless of any other activity. If that's
> true, then it's the OS's fault that it can't keep time, since the computer
> itself (RTC) always knows the correct time without any appreciable losses.

yes, but the granularity of the rtc isn't high enough for unix purposes.
we could probably use one of the unused timer interrupts (i know, there
are other issues with this) to schedule reads of the rtc under netbsd and
thus keep better time, but the problem then is that the clock tends to
skip forwards in a rather unexpected way.

> Personally, I don't mind that the clock was done that way...it makes sure
> that the clock function doesn't slow any other interrupts, but still lets
> the computer keep accurate time, given that it's simply "wrong" not to
> reboot a computer to clean up its RAM.

what if you're running a server that should never go down?  if the mac
memory management scheme weren't so screwy, this wouldn't be quite as
necessary.  i realize that the mac scheme is a valid way of doing things,
but in general, i find the unix programming model for handling memory to
be far more intuitive (handles just seem a bit nuts to me ;-)

> I realize that I'm in the minority on this issue, and knew it when I first
> wrote, but again, as I said, we're an unauthorized use of the Mac hardware,
> so the NetBSD opinion isn't really going to count much in future design
> decisions. But...on the bright side, MacOSX will be unix-based, so it may
> receive an overhaul anyway (I dunno anything about PPC/G3 implementations).

well, i don't think we're hoping to change anything, just bitching and
moaning about how things currently stand...completely useless, of course
;-)

> Nothing like a totally moot conversation that is meaningless since it'll
> never accomplish anything ;-)

of course!!!!

later.

-- 
Colin Wood                                 cwood@ichips.intel.com
Component Design Engineer - PMD                 Intel Corporation
-----------------------------------------------------------------
I speak only on my own behalf, not for my employer.