Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.programming > #3481 > unrolled thread

Re: Dmitry Vyukov's distributed rwlock 2

Back to article view | Back to comp.programming

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: Dmitry Vyukov's distributed rwlock 2 "Chris M. Thomasson" <no@spam.invalid> - 2013-06-20 13:26 -0700
    Re: Dmitry Vyukov's distributed rwlock 2 "Chris M. Thomasson" <no@spam.invalid> - 2013-06-20 14:01 -0700

#3481 — Re: Dmitry Vyukov's distributed rwlock 2

"aminer"  wrote in message news:kpvj65$cuq$11@dont-email.me... 

On 6/20/2013 2:51 PM, aminer wrote:
> > Dmitry Vyukov's asymmetric rwmutex algorithm is not NUMA aware,
> > so i prefer to use Dmitry Vyukov's distributed rwlock combined
> > with my RWLock algorithm.



> Sorry, the Dmitry Vyukov's asymmetric rwmutex algorithm is scalable
> on the reader side on NUMA also, but i will use the Dmitry Vyukov's 
> distributed rwlock combined with my RWLock algorithm.

Very Good Aminer!

:^D

Now, THINK about _combining_ asymmetric synchronization 
with very clever distribution techniques...

[toc] | [next] | [standalone]

#3483

> "aminer"  wrote in message news:51C395BC.4020204@toto.net... 

> On 6/20/2013 1:26 PM, Chris M. Thomasson wrote:
> > Very Good Aminer!
> >
> > :^D
> > > 
> Now, THINK about _combining_ asymmetric synchronization with very clever
> > distribution techniques...

> I have read on internet about asymmetric synchronization , but
> they have not explain it easily, the documentation on asymmetric 
> synchronizationis is still difficult, but anyway, if i combine
> my RWLock and the distributed RWLock this will be very good also,
> cause this will eliminates the contention on the reader side and 
> eleminates the hotspot on the reader side so there will be just
> a small gain in speed if you combine a distributed RWLock with and 
> asymmetric synchronization.

Ahhh, IMVHO, I think you might be "missing" the fact that one can use 
atomic RMW on the writer side, and virtually zero sync on the reader.
Therefore, you distribute the writer side only, and allow readers to be
atomic RMW free for asymmetric synchronization allows for this setup...

[toc] | [prev] | [standalone]

Back to top | Article view | comp.programming

csiph-web