tut@ucbopal.CC.Berkeley.ARPA (06/17/84)
One of the slides for Gene Dronek's talk in Salt Lake City on benchmarking showed that 3B2 performance was less than half as good as a decent MC68000 system (such as Dual) running Uniplus. People may have missed this, as it was early Friday morning. Does anybody who has used a 3B2 have any comments on this? The one I saw at the show seemed fast enough, though I didn't like the unavailability of tape backup. Bill Tuthill
eric@aplvax.UUCP (06/20/84)
Some early results from running the Uniq benchmarks indicates that the 3B2 is about half the speed of a good 68000 machine (Plexus). The slow-down in speed seems to be primarily due to a slow disk. Once we get a compiler ( :-) ), I'll run some more benchmarks on it to test general cpu and OS speed. -- eric ...!seismo!umcp-cs!aplvax!eric
bytebug@pertec.UUCP (roger long) (06/23/84)
Isn't the 3B2 the one that AT&T says will support 18 users for $10K?
agk@ihuxq.UUCP (Andy Kegel) (06/26/84)
> Isn't the 3B2 the one that AT&T says will support 18 users for $10K?
It is, and it does, but THE USUAL CAVEATS APPLY.
We have a ("1 MIP") machine around here that supports one user -- the
person who builds the development UN*X kernel each day. UC-Berkeley
has (had) a PDP-11/70 that supported some HUGE number of users (the
rumor was 70 simultaneous users).
The Point: the number of users is a STRONG function of what they do.
The AT&T 3B2 system will not reasonably support 18 software developers
doing simultaneous "cc kernel.c" commands. On the other hand, the
machine will reasonably support lots of users in an office environment
(a mixture of spreadsheets, data base activity, word processing,
e-mail, and games -- note tongue firmly in cheek).
The above has been vague for two reasons:
1) Others have obtained the results I summarized. I feel *they* have the
right to announce the results of their work, not I.
2) Learning to be a good competitor means learning when to speak, and when
to be quiet. This is a hard lesson to learn. Please bear with me.
A useful discussion to derive from this would concern benchmarks. Does
compiling 'printf("hello, world");' accurately measure a machine? Do
the Whetstone benchmark speeds accurately model your floating point
needs? Does your computer spend a lot of its time in Ackerman's function
or solving the now-famous "puzzle?" Does anyone have a spell(1) dictionary
that includes "Ackerman" and "Whetstone" correctly spelled?
The preceeding represents my personal opinions, and has not been cleared
nor approved by the Company, the NSA, the KGB, nor my mother.
-andy kegel