Re: [m2t-dev] Backend documentation

[Cédric Brun <cedric.brun@xxxxxxx>]

Hi Arno,

I agree with your point about the intrinsic complexity of handling threads in 
the backend layer and that this decision can be addressed at a higher level 
of abstraction.

Do any of the M2T component have multi-threading features yet ?

On another side we'll start evaluating the backend soon starting on what have 
been done in eXtend and XPand, we still have concern about how well we'll be 
able to map OCL expression (as MTL is strongly based on every aspect of OCL) 
to the backend. I guess we'll have many questions then :) 

XPand is using a subset of OCL, isn't it ? Is it using the Eclipse OCL plugin 
or not at all ?

Cédric

Le Monday 28 January 2008 17:38:14 Arno Haase, vous avez écrit :
> Hi Cedric,
>
> thank you for your reply. If you have further questions from looking at
> the source, do not hesitate to ask!
>
> With regard to thread awareness, we looked into it but decided against
> incorporating support into the backend. Our rationale for this decision
> is that fine grained interaction at the execution level would entail a
> huge amount of synchronization overhead (both technical "synchronized"
> sections and logic to have one part wait for another).
>
> We think that many scenarios can be split into several parallel threads
> of execution with little to none interaction, e.g. generating a lot of
> independent artifacts. But we think that in those cases, parallelization
> at the workflow level - i.e. execute several workflow components in
> parallel - is a far better choice because it really isolates the
> execution, it has a far simpler thread model, it is a high level design
> decision that can be addressed at a high level of abstraction.
>
> But since you raised the question, maybe now is a good time to discuss
> this. Any opinions?
>
> - Arno
>
> Cédric Brun schrieb:
> > Hi Arno,
> >
> > Thank you for the doc, I went through it and I'm now looking at the
> > source code. It seems really nice and we'll study it more closely to use
> > it in MTL.
> >
> > I am wondering about multi-threading, is there some work there to ease
> > the creation of a multi-threads generation when "executing" the data
> > structures of the backend ?
> > I had not been able to find something in the source code yet so I would
> > be glad if you enlighten me about that. I did not conduced experiments
> > neither but as multi-core CPUs are quite common these days I guess the
> > backend should be "thread-aware", shouldn't it ?
> >
> > If anybody have figures about the IO versus CPU usage during a generation
> > for a common use case, it could help starting the debate too :)
> >
> > Cédric
> >
> > Le Monday 28 January 2008 15:16:47 Arno Haase, vous avez écrit :
> >> Hi everyone,
> >>
> >> it has been a while since I put the backend documentation up in the
> >> Wiki, and so far I have received no reactions (except for two of the oAW
> >> guys who are collaborating closely with me).
> >>
> >> It would be really helpful for me to get feedback on the ideas as early
> >> as possible - especially from you JET and MOF Script guys - so that I
> >> can accomodate your needs!
> >>
> >> Hope to hear from you soon!
> >>
> >> - Arno
> >
> > _______________________________________________
> > m2t-dev mailing list
> > m2t-dev@xxxxxxxxxxx
> > https://dev.eclipse.org/mailman/listinfo/m2t-dev