Re: clisp MT & summer of code

[Vladimir T. <vtz...@gm...>]

IMO, the only major thing left to be implemented are thread safe hash tables.
There are two approaches to accomplish this:
1. have lock on hash tables operations (e.g. add :synchronized param
to make-hash-table and ensure no segfault may happen with or without
it).
2. re-implement entirely hash tables.

(1) while easier is not good according to me. This will cause
performance issues in clos (internally it uses hash tables) - even
make-instance should (most probably) obtain lock internally.
(2) is the way to go, IMO. I think we should replace current hash
tables with lock-free open-addressing ones like described in:
http://www.azulsystems.com/events/javaone_2007/2007_LockFreeHash.pdf.
CCL hash tables implementation is based on this as well.
The hardest part in this reimplementation is integration with GC
because of weak relations.
I've started to implement (2) but in last year I had almost no free
time to spend on it :(.

There will be more issues with mt of course - but not so major ones as
hash tables.

Vladimir

PS:  currently there is problem with generational gc and heap holes
filling - hope this week to fix it. the problem can be easily observed
when some threads perform socket i/o from slow server and other worker
threads trigger GC frequently.


On 2/13/11, Sam Steingold <sd...@gn...> wrote:
> Vladimir,
> I think we should create a clisp/mt project for the google Summer of Lisp.
> IIUC, the only remaining hurdle is MT-safe hash tables, and this sounds
> like a good student project.
> Could you please write a brief description of what needs to be done?
> Thanks!
> --
> Sam Steingold (http://sds.podval.org/) on Ubuntu 10.04 (lucid)
> http://memri.org http://thereligionofpeace.com http://mideasttruth.com
> http://truepeace.org http://iris.org.il http://pmw.org.il http://camera.org
> XFM: Exit file manager? [Continue] [Cancel] [Abort]
>