[Shootout-list] process creation & message passing chain
Bengt Kleberg
bengt.kleberg@ericsson.com
Sat, 06 Nov 2004 14:42:18 +0100
first i would like to thank isaac for posting a about-this-test for
these 2 tests. for me, that do not understand the motive/reasoning
behind a test design by reading the code that implements it, it has been
invaluable.
Isaac Gouy wrote:
...deleted
> We measure process creation sending just one message (the count).
unless we worry about the (hypothetical) language where message passing
is so slow that even one message will influence the process creation time.
if we worry, an alternative would be to let each process prove its
existence by creating the next process. thus no message passing would be
needed.
...deleted
> In chainmsg the process creation / message passing ratio can be
> adjusted to satisfy Bengt.
presumably it would satisfy all those that do not know how many messages
that are needed to sufficiently minimise ''the process creation time''
influence on the messsage passing test.
> We still have to establish that on the test machine, the fast languages
> have measurable times without requiring such a large N (number of
> processes/threads) that other languages suffer "death by concurrency"
> (to satisfy me).
how about a 3;rd argument (T?). it could be used to loop over the test
''body'' T times, so that the total time for a test would be
(time-for-1-test)*T. T would have to be trial-and-errored until the
shootout machine/os gives us a measurable time.
(i still have not received an answer to my question about how much time
this is. therefore i have sofar ignored the problem).
> In addition, using the chain programs to measure the maximum number of
> processes/threads that can be created on the test machine, and
> including those limits in the 'about' text seems fine to me.
surely it would be better to use a graph? the shootout have graphs, but
they are underutilised at present.
bengt