[SunRescue] SS4/SS5/SS10 funzies.....(:+}}...

[James Lockwood]

On Tue, 3 Apr 2001, Dave McGuire wrote:

>   James, I've learned not to argue with you on technical points, but
> on this one I must disagree.  While [again] I haven't looked at the
> benchmarks, in real-world apps, SS10s with SM81s (or even SM71s)
> have beaten the crap out of SS5/170 machines for me.  YMMV, I guess,
> but that's what I've seen, over and over.  Maybe I've been getting
> lucky cache-wise or something, but again, that's what I've seen.

If your code benefits from extra cache, the Supers will win.
If your system context switches heavily, the Supers will win.
If your code is FP intensive, the Supers will win.

Not surprisingly, these 3 factors are usually exist simultaneously in many
interactive systems.  I would rather have a SS20/81 in front of me than a
SS5/170 if I were going to run X on it, the speed boost is substantial.
Add to that the fact that an SX framebuffer is much nicer than the S24 and
you have a pretty nice package.

For server duty it's a bit more even.  Benchmarks of some of my code
(semantic document analysis, int heavy long lived single processes) showed
them within 5% of each other.  On the other hand, the SS20 can hold twice
as much RAM and more processors, so for true server applications where you
can afford to expand the box to the moon the 20 will win.

As always, YMMV.

-James