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

Groups > comp.programming.threads > #2219 > unrolled thread

More information...

Back to article view | Back to comp.programming.threads

Contents

  More information... aminer <aminer@toto.net> - 2014-04-22 19:26 -0700
    Re: More information... aminer <aminer@toto.net> - 2014-04-22 19:31 -0700

#2219 — More information...

Hello,


You have to know that a TicketSpinlock with a proportional backoff
has a problem , if you use 4 threads on 4 cores the Ticketspinlock
will do very well its job, but if you use more threads than the
number of cores the TicketSpinlock will not scale, this problem
do not happen with a simple Spinlock with a backoff , a simple
Spinlock with a backoff will scale beautifully even if the number of 
threads is greater than the number of cores, that's why i have
used a simple Spinlock with a backoff inside my SemaMonitor and used
it inside this fast concurrent FIFO queue that is giving
a throughput of 6.4 millions transactions per second even if i am using 
my SemaMonitor with it, just test it yourself and you will notice
that it's very fast, and the pushed and the poped items will still be 
done in a FIFO order, so i think that this fast concurrent FIFO queue is 
great.


You can download this fast concurrent FIFO queue from:

http://pages.videotron.com/aminer/



Thank you,
Amine Moulay Ramdane.

[toc] | [next] | [standalone]

#2220

On 4/22/2014 7:26 PM, aminer wrote:
> You have to know that a TicketSpinlock with a proportional backoff
> has a problem , if you use 4 threads on 4 cores the Ticketspinlock
> will do very well its job, but if you use more threads than the
> number of cores the TicketSpinlock will not scale,


This problem do happen also with an array based lock or queue based lock 
such as MCS and CLH.

So the simple Spinlock with a backoff scales well and avoids this 
problem and that's very important to know.


Thank you,
Amine Moulay Ramdane.

[toc] | [prev] | [standalone]

Back to top | Article view | comp.programming.threads

csiph-web