[Shootout-list] Re: ring of processes

Aaron Denney wnoise@ofb.net
Tue, 12 Oct 2004 16:24:25 +0000 (UTC)


On 2004-10-12, Bengt Kleberg <bengt.kleberg@ericsson.com> wrote:
> i had in the beginning a fixed number of processes, and variable number 
> of messages. however, the fixed number of processes made the test into 
> one of ''create processes'', and not ''send messages'', for small number 
> of messages. i think my current wording makes the potential problem more 
> visible.

True.  I'd say "then don't use small Ns".  Many of the tests have some
significant startup cost that is absorbed by using large N.

> does everybody understand that ''128 messages'' is per processs? we do 
> have a variable number of messages in the test. it is just the ratio of 
> messages to processes that is constant.

Yes, that was clear enough to me that I didn't feel any need to
emphasize it when responding.  One concern I do have is that it may
make large N values difficult to test; running with high N may create
too many processes under implementations that map execution contexts to
processes.

Oh, and calling it "ring message" seems to overemphasize one link used
between the sink and the source.  "Line message" seems a much better
description.

(Not to nitpick more, but really, the sorts of scenarios that these IPC
tests seemed designed to simulate could be solved in Haskell by lazy
lists.  My prime number directed-message test couldn't easily, which
is why I'd still like to see something like it go in.)

-- 
Aaron Denney
-><-