oster (09/21/82)
**************************************************
This message is not for publication.
**************************************************
When I was at the Usenix conference, over July 6-9. in
Boston, Ma., a benchmark was going around.
The guy doing the benchmarking kindly let me borrow his figures.
These are the results:
____________________________________________________________
Performance of Some Micro Systems.
Computer compile time (sec) run time (sec)
run user system run user system
Zilog Z-8000 38. 6.4 6.2 49. 25.6 9.6
Onyx c8002 wouldn't compile
Plexus-24 22. 9.5 4.6 1:03. 36.4 20.6
Pixel 52. 10. 7. 3:21. 2:36.2 12.8
Dual 36. 13.6 7.7 4:16. 3:43. 13.7
Callan 53. 18.3 7.7 5:30. 3:37.8 15.8
Altos 8086 38. 9.4 10.9 1:50. 59.9 34.4
Vax 11/780 wi 18 users 1:46. 14. 6.1
Codata 45. 18.4 9.3 5:21. 4:48. 14.5
Fortune 32:16 31 5.7 3.4 wouldn't run
____________________________________________________________
The benchmark created some 20k long files, then called sort and join a few
times.
(I'm sorry, I don't have the text of the benchmark.) I watched the
benchmark being run on the Fortune system. The benchmark failed when
it came to the step:
system("sort afil -o bfil");
where afil had 40000 characters in it. (Sort printed an error message and
the benchmark died. The operating system didn't crash.) Earlier in the
benchmark it had sorted a 20k file. This is disturbing because sort is
supposed to use temporary files if its input is to big to fit in main
memory. This is doubly disturbing when you consider that you have to pay
extra for what I am used to thinking of as the standard Unix utilities.
(Do you have to pay extra for this sort?)
As far as I'm concerned, if I don't get the full set of utilities, then
I'm not getting Unix.
David Oster
lime!burdvax!oster
(Incidentally, I discovered a bug in the IBM PC running Coherent -
when I tried to use evaluating-quotes in the shell
$ ls -l `echo a*`
It just gave me another shell prompt - it didn't run the outer command)