Cell Relay Archive

[vjs@calcite.rhyolite.com (Vernon Schryver)]

In article <3lssua$mvk@hpindda.cup.hp.com> raj@cup.hp.com writes:
>Vernon Schryver (vjs@calcite.rhyolite.com) wrote:
>: Besides, it takes time to port and run benchmarks.
>
>I trust that you are not trying to say that porting a benchmark which
>can measure latency such as netperf is truly a time consuming task?!?
>I know that it has run on the Unix biggies (DEC, HP, IBM, Sun, SGI)
>for at least two years. After all, it's not as though this is
>something big and complicated like TPC or SPEC SFS.


I figure it would take most of a day to get some of the numbers you've
been nagging me about, after I managed to find the right kinds of hardware
running a kernel that is not ridiculous (e.g. compiled with debugging
so that ttcp numbers are reduced by 50%).  After getting the numbers,
you must understand them well enough to ensure they're not so bogus you
get in trouble for publicizing them.

My trouble is that if I had spare half a day, I'd spend it doing things
like enabling keepalives and port-number-reusing in ttcp to make it
easier to run as a "soaker" on new hardware that likes to lock up.

Also, if people caught me running benchmarks, they might think that's
my job, and expect me to do it regularly.  Anyone who has been stuck
with that for more than 10 minutes quickly loses the ability to type,
hear the names of benchmarks, or something, anything to be disqualified.


Vernon Schryver    vjs@rhyolite.com