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

Groups > comp.programming.threads > #2144

Re: To be more precise...

Cross-posted to 2 groups.

Show all headers | View raw

On 4/11/2014 8:14 PM, aminer wrote:
>
> Hello,
>
>
> We have to be smart please, as you have noticed i have
> wrote another scalable FIFO fair lock and i have
> studied deeply , and i have come to the conclusion
> that it's useful to learn and useful to use, i have
> began to test it under contention to see how it behave,
> under contention with 2 threads on a Quadcore and
> under contention with 4 threads, what i have
> discovered that under contention with two threads
> the following section in the Enter() method
> will be accessed by 1/2 of the total number of threads

I mean 1/2 the total number of times the threads enter the Enter()
method.


>
> here is the section, it is when PMyRecord4(a)^.count equal -1:
>
> if PMyRecord4(a)^.myid=0
>            then
>             begin
>              PMyRecord4(a)^.count:=1;
>              exit;
>              end
>           else
>           begin
>            freemem(PMyRecord4(a)^.mem); <- this one
>            end;
>        end
>
> but under contention with 4 threads this same section
> will be accessed almost 0 times.. so if you want to test
> the scalability with  two threads scenario under contention,
> you will not be able to do it accuratly cause under contention
> with two threads the queue.pop(tobject(a)) statement inside the Enter()
> method will be cheap compared to the 4 threads scenario under
> contention, but what i can tell you for sure is that with 4 threads and
> more and with more cores and under contention my FIFO fair lock
> will be scalable.
>
> But we have to be smart more than that, you will say that my
> scalable FIFO fair lock is slower by almost 2x times than the
> Ticket spinlock with a proportional backoff, but you have
> to understand that the TicketSpinlock will not scale with
> bigger critical sections ,  my scalable FIFO fair lock
> will, other than that you have to understand that as the time
> inside the  critical section becomes bigger the speed
> of my scalable FIFO fair lock will be almost equal to that of
> the TicketSpinlock with a proportional backoff, other than
> that, to make my scalable FIFO fair lock faster you can use
> a Waitfree FIFO queue on the pop side of the queue.
>
> You can download my scalable FIFO fair lock from:
>
> http://pages.videotron.com/aminer/
>
>
>
>
> Thank you,
> Amine Moulay Ramdane.
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>

Back to comp.programming.threads | Previous | Next — Previous in thread | Find similar | Unroll thread

Thread

To be more precise... aminer <aminer@toto.net> - 2014-04-11 20:14 -0700
  Re: To be more precise... aminer <aminer@toto.net> - 2014-04-11 20:17 -0700

csiph-web