[gambit-list] crash on callback from WMI event

Marc Feeley feeley at iro.umontreal.ca
Wed Jun 21 15:59:34 EDT 2006


On 21-Jun-06, at 1:46 PM, Logan, Patrick D wrote:

> Marc,
>
> The recent Erlang release now supports SMP.
> (http://erlang.org/doc/doc-5.5/doc/highlights.html)
>
> The doc says it runs a "process scheduler" per OS thread or per actual
> processor. (Not sure which it actually enforces.) I assume "process
> scheduler" amounts to "virtual machine" or enough of the VM to use
> thread-specific data.
>
> Is a conversion of Gambit to something like this a really big or  
> just a
> fairly big effort?  8^)
>
> -Patrick

I estimate the amount of work to about one month of my time (full- 
time).  That's a lot of work.  I could probably get a first version  
working quickly, but with parallelism its the last few bugs that take  
90% of the time.  There are many critical sections in the code that  
would need real mutexes in an SMP version.  Currently I take  
advantage of the fact that these mutexes are not needed on  
monoprocessors (this has implications for development time, i.e. I  
have to identify all the critical sections, but also in terms of run  
time, i.e. some operations will become slower in the SMP version  
unless some costly wizardry is used).

Currently SMP is not at the top of my TODO list.  I want to get the  
module distribution system working first and improve the performance  
of code with no declarations (pure R5RS Scheme code).

Marc




More information about the Gambit-list mailing list