[plt-scheme] Threading vs. System Threads

From: Michael Sperber [Mr. Preprocessor] (sperber at informatik.uni-tuebingen.de)
Date: Wed Feb 12 04:24:30 EST 2003

>>>>> "Brent" == Brent Fulgham <bfulgham at debian.org> writes:

Brent> The conjecture is that the use of OS-level threads for concurrency leads
Brent> to lower performance than Erlang's internal concurrency.  I like to imagine
Brent> that MzScheme may share some of these positive features (Matthew,
Brent> please feel free to correct me! :-)

This argument assumes that the only way to "have native threads" would
be to map language-level threads one-to-one onto OS threads.  For this
situation (which, I believe, reflects those older versions of MzScheme
which did "do native threads"), of course your incur the (often
substantial) overhead and limitations of OS threads in your
language-level threads.

This is mainly because you're using a mechanism for a purpose it
wasn't really designed for: OS-level threads are usually for
exploiting machine-level parallelism, language-level threads are most
useful for enabling certain program paradigms.  Both have wildly
different requirements.  Therefore, it can make perfect sense to run
language-level threads on multiple OS threads, but (at least in the
case of Scheme), to reap the benefits, you really want to run many
language-level threads on a small or even fixed number of OS threads.

Thus, the absence of using OS-level threads to (help) implemement
language-level threads by itself doesn't mean anything, positive or
negative.  Having said that, I'm pretty sure MzScheme threads have
substantially lower overheads than your typical OS threads

Cheers =8-} Mike
Friede, Völkerverständigung und überhaupt blabla

Posted on the users mailing list.