[net.unix-wizards] Sun-3 benchmarks

jqj@cornell.UUCP (J Q Johnson) (10/15/85)

Does anyone have a set of benchmarks for the new Sun-3 yet?  I'm aware of
the Spanier paper which claims roughly twice a uVAX II in raw cpu, but
that paper has enough problems (ambiguities, inconsistencies, and suboptimal
choices of non-SMI configurations) that I am quite suspicious of its results.

What I'm looking for is a set of benchmarks, including raw data and programs,
that purport to measure system throughput in a typical multiuser Unix
setting.  I would be particularly interested in seeing the Sun-3 with
SMD/Eagle disk and 4 or 8MB memory compared to a uVAX II with Emulex (or
SI)/Eagle disk and 5 or 9MB memory.  I am less interested in raw cpu
benchmarks, but I would be interested in benchmarks of particular components
of general system throughput.

As is customary, if people reply to me directly I'll summarize any
interesting responses.

meissner@rtp47.UUCP (Michael Meissner) (10/18/85)

In article <1171@cornell.UUCP> jqj@cornell.UUCP (J Q Johnson) writes:
>
>What I'm looking for is a set of benchmarks, including raw data and programs,
>that purport to measure system throughput in a typical multiuser Unix
>setting.  I would be particularly interested in seeing the Sun-3 with
>SMD/Eagle disk and 4 or 8MB memory compared to a uVAX II with Emulex (or
>SI)/Eagle disk and 5 or 9MB memory.  I am less interested in raw cpu
>benchmarks, but I would be interested in benchmarks of particular components
>of general system throughput.
>
I recently came across one of the better UNIX benchmarks, in helping an OEM
with a bid for the federal court system.  I don't know whether the benchmarks
are the property of the government or not, but they seemed rather thorough.
They were for System V (as opposed to 4.2).  Things tested include:

	CPU speed (whetstone and dhrystone)
	performance of the disk (creating a 100M file and doing random I/O)
	performance of the terminals (whether multiple terminals could
		simultaneously output 9600 baud)
	performance of nroff
	Paging performance
	Multiuser performance (various of the above things at the same time)