[self-interest] Self & Parallelism

[Jecel Assumpcao Jr]

Anne C. van Rossum wrote on Wed, 24 Jan 2007 10:29:07 -0000
> Online I read some articles about Self. I find it a very interesting
> language. I'm also interested in the optimizing compiler, but with a
> slightly different touch.

Welcome to Self! Sorry that your email got stuck in moderation for two
weeks - I had some problems with my spam filter eliminating Yahoo's
notification message.
 
> I'm thinking about a way in which an application can be sliced into
> different tasks. Within a task entities may communicate with each
> other in a synchronous way. Between tasks communication is in a
> asynchronous way. (A task is a threaded entity.) An optimizing
> compiler dynamically reconfigures the application in an amount of
> tasks that performs optimal. This depends e.g. on the amount of
> processors.

After the 1980s interest in this topic declined significantly, but now
it is more popular than ever.
 
> I'm interested how parallelism, concurrency, distribution, etc. is
> seen in Self. What kind of abstractions are used or invented? Are
> threads adopted? Or are communicating sequential processes used? Or
> does Self have its own view on concurrency too?

Self has the same model as Smalltalk-80, which is threads with fixed
priorities that synchronize using semaphores. Unlike most Smalltalk
implementations where this is hidden inside the virtual machine, in Self
the scheduler is just a regular program that you can change or replace.
Reinaldo Silveira, for example, took advantage of this to implement a
very different concurrency model for his PhD thesis.

You might find the work I did with tinySelf 1 interesting, even though I
implemented a whole new interpreter (in Self) instead of just changing
the scheduler:
http://www.lsi.usp.br/~jecel/tiny.html#rel1

There is also this paper that I wrote (compressed postscript rather than
PDF - sorry) where I speculate about the use of the adaptive compilation
technologies to adjust the level of parallelism in the running
application to the number of processors available:

http://www.lsi.usp.br/~jecel/jabs1.html

-- Jecel