Getting locks and sharing: was RE: Getting concurrency

[Thomas Lindgren]

--- Reto Kramer <kramer@REDACTED> wrote:
> I find it very unnatural to have to conceive of
> multiple nodes per 
> machine "just" because there's no VM technology in
> Erlang that takes 
> advantage of multi-CPU machines _transparently_.

Note that an approach of using multiple sequential
nodes instead of a multithreaded node could be fairly
transparent to the programmer, with the appropriate
high-level support.

Also note that it's not clear how transparent the
issues of parallelism _should_ be, in either approach.

> Where can I learn about what's been done in the past
> and get a sense of 
> the Erlang specific challenges and the complexity of
> such a multi CPU 
> aware VM?

Have a look at Pekka's MSc, or dig your way through
the endless ranks of papers on implicit
parallelization from the 80s to the early 90s :-)

I'd say the first issue for a multithreaded VM is to
make the runtime system thread safe (and robust). The
second issue is to make it scalable. The third issue
is to make it competitive with a sequential VM.

(And somewhere in that list, we can begin worrying
about how well our whole erlang systems perform :-)

Best,
Thomas


__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com