Subject: Re: ballpark ratio users:cpu for a 86 NetBSD box?
To: Dave McGuire <firstname.lastname@example.org>
From: Perry E. Metzger <email@example.com>
Date: 09/27/1994 16:16:45
Dave McGuire says:
> "Better in every respect" is flame bait that I will ignore.
The machines have
1) Faster IO
2) Faster context switching
3) Faster processors
So whats left to be valuable?
> "Far more powerful" is true in some respects. However, FOR OUR
> APPLICATION, ( <-reread that) FOR OUR APPLICATION, Sparcstations, even
> stuff like 10's, simply won't handle the users.
Thats silly. If a 10 won't handle it, then a 4/400 series machine with
far fewer contexts, a slower bus, slower processors and the like is
certainly not going to handle it.
> We have a major competitor that tried it (two, actually) and it bit
> them on the butt.
Probably bad engineering, not bad machines.
> Engineering workstations are simply not designed to replace big
> multiuser systems.
What does that mean? What, concretely, is missing? The IO is fine. The
speed is fine. The contexts are fine. The OS is fine. Whats left to be
"not designed to replace big multiuser systems"?
> About the "FOR OUR APPLICATION" stuff... I stress this because I
> have no doubt that the stack of Sparcstations were better in *their*
> application, but MY application here is the only one for which I can
> speak of as having a great deal of experience with. I would not
> presume to debate you in *your* area of expertise.
If your application is IO bound, well, the sparcstations have faster
IO. So does the Pentium running PCI and decent controllers, drives and
If your application is context bound, well, the same story goes.
If your application is memory bound, the newer machines can handle
more memory and faster memory.
If your application is processor speed is the problem, the newer
machines have bigger caches and faster execution times.
So what is the explanation you can give for why the old stuff is
better for you?