Subject: Re: probing CPU speed?
To: Jonathan Stone <jonathan@DSG.Stanford.EDU>
From: Chris G. Demetriou <>
List: port-i386
Date: 11/19/1998 09:04:45
Jonathan Stone <jonathan@DSG.Stanford.EDU> writes:
> I can think of 3 plausible, legitimate uses without even blinking:
>     a) mapping cycle-counter measurements of *real* code
>        execution to wall time.
>     b) calibrating use of cpu cycle-counter as NTP  clock source.
>     c) ascertaining whether one of several externally-identical
>        machines has been re-clocked or overclocked.
>        say, in a lab.  These things happen...
> [ ... ]
> so, if someone really does want to know the CPU clock rate for
> something useful, why not let them? 

I'm perfectly willing to say that "if you have a cycle counter and can
easily print its frequency, or if you easily know your CPU speed,
print it."

In fact, I _DID_ that when i wrote the alpha code.  It prints CPU
speed, as reported by the Alpha HWRPB cycle counter frequency, on
system startup.

However, that's a far cry from a "BogoMIPS"-like measure, which is
what was suggested and what i was replying to.  (Note that that is the
message i replied to and quoted, not your previous messages.!)

How do your uses apply to that?  I don't see that they do.

> Arguing that `I dont see a legitimate use for this, thererefore you
> cannot have any legitimate use for it'-- well, do we really want to be
> that arrogant?
> If you dont want it to be misused, fine, make it an option or something.

We (the developers of NetBSD) are in the operating system business.
It's our _job_ to decide what kinds of things are legitimate and
reasonable to put into our kernel, based on our own personal and
collective judgement.  As soon as we decide that we can't exercise
that judgment, we'd do quite well to exit the kernel business.

So, again, what real benefit does a "BogoMIPS"-like metric have?

I've heard one suggested: you know that if it _changes_, something odd
has happened.  However, isn't that also true, if, say, you run the
'mhz' program on a quiet system?  Why do you need to add this feature
to the kernel, to all ports (as was suggested; cross-platform
CPU "performance" measurement was given as a potential use), to do

Chris Demetriou - -
Disclaimer: Not speaking for NetBSD, just expressing my own opinion.