Real Time Performance and Networking (was: re: Faith of Correspondents?)

Dean_Swan at Mitel.COM Dean_Swan at Mitel.COM
Mon Nov 8 17:26:20 UTC 1999



From:  Dean Swan at MITEL on 11/08/99 12:26 PM

Bijan Parsia wrote:

>Right now, it's *impossible* to write a web server/web application
>that can handle multiple simlultanous requests (at least, on the mac).

[snip]

>Add that in an ideal world (;)) the same image might also be running a
>ftp server, or a mail server, or be used as a client by someone, and
>things get a bit messy.


I, for one, am anxious to take a look at Tim's paper regariding real-time
Squeak.  I think one of the things that certainly contributes to the performance
(or lack thereof) for server side networking applications in Squeak is the lack
of fully pre-emptive scheduling.

As things currently stand, it's *VERY* easy to have a single process seize the
Processor for extended periods of time, unless one is very careful.

I'd like to see Squeak move a little bit in the direction of the process model
of Smalltalk/X.  Smalltalk/X seems to be more developed with respect to
supporting multi-tasking/multi-threading (i.e. every window running in a
separate process, full pre-emption support, thread-safe core classes).

I also think that Morphic would benefit HUGELY from better scheduling policies
and capabilities.  As things currently are, the Morphic kernel in
PasteUpMorph/DoOneCycle is essentially simulating a round-robin scheduler, and I
really think it makes more sense to let ProcessorScheduler do the heavy lifting
here.  This would also allow changes in scheduling policy to be focused in one
place, and make things like time based animation much simpler.

I would like to hear what others think about this, especially anyone from SC.


                              -Dean Swan
                              dean_swan at mitel.com





More information about the Squeak-dev mailing list