Subj : Re: Challenge: Multithreading & Synchronization
To   : comp.programming.threads
From : Uenal Mutlu
Date : Thu May 19 2005 07:21 pm

"Sergei Organov" wrote
> "Uenal Mutlu" writes:
> > "Sergei Organov" wrote
> > > "Uenal Mutlu" writes:
> > > > > "Uenal Mutlu" writes:
> > > > >
> > > > > [...]
> > > > > > But if the data objects are independent of each other then it is more
> > > > > >efficient to let each have its own mutex.
> > > [...]
> > > > In this case IMO it does not make much sense to lock X, Y, Z
> > > > individually. It depends on what kind of access to the data you
> > > > want for your threads. It's an issue of the application logic.
> > >
> > > In the above I've skipped everything and left two consecutive statements of
> > > yours. Please read them... Done?
> > >
> > > Now I wonder are you actually one person or at least two?! Hopefully you
> > >don't in fact experience split personality :)
> >
> > No, I don't. But try to understand that these things are complex and
> > difficult to describe.
>
> > You on the other side are looking for simple answers to such complex
> > problems.
>
> No, I don't. It's you who are looking for simple solution to a complex
> problem and believe that recursive mutex is the answer, isn't it?

Yes, I try to find a way to simplify, to generalize these complex things,
if possible to find an algebra for it.

> > Try to understand my reply to Maciej Sobczak. There I've tried to
> > explain these cases along with examples. The right decision depends on
> > the access level to the data you want to grant each thread, and the
> > concurreny level among the threads.
>
> I didn't have any objection. What I'm saying is that your own statement
> cited above:
>
> > > > > > But if the data objects are independent of each other then it
> > > > > > is more efficient to let each have its own mutex.
>
> now appears to be plain wrong from your own point of view, right?

Hmm, sure it does not apply to such extreme examples like your 3D-point,
but for most practical data (esp. containers/arrays) which have to be accessed
concurrently it does, doesn't it?
I have to wonder how you come on such an idea to believe that my statement
would include even such trivial cases like the 3 parts of a 3D point, hmm...

.