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

Groups > comp.programming.threads > #2285 > unrolled thread

Follow with me please...

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

Contents

  Follow with me please... aminer <aminer@toto.net> - 2014-04-28 00:35 -0700
    Re: Follow with me please... aminer <aminer@toto.net> - 2014-04-28 11:35 -0700

#2285 — Follow with me please...

Hello,

Follow with me please...


I have just explained to you why the following Chriss Thomasson
algorithm:

https://groups.google.com/d/topic/lock-free/acjQ3-89abE/discussion

has scored 33% more throughput on the pop() side than my algorithm..

Here is my algorithm:

http://pages.videotron.com/aminer/CQueue2.htm


So follow with me please cause we have to be smart to understand
the inside of those scalable algorithms...

I have just said that since i am using more variables that generate
data movements between caches and between the memory subsystem and the 
caches, those data movements causes a contention on the "Bus" system
and the Bus system must serialize those data movements, and more than
that since there is contention in this scenario as i have just explained 
this "factor" that we call "contention" in this scenario
is highing the waiting time of the threads and is lowering the 
throughput,so since i am using 4 variables that generate data mouvements 
between caches and between the memory system and the caches
and since the Chriss Thomasson is using 3 variables, so my algorithm
is causing more contention on the Bus than the Chriss Thomasson 
algorithm and this explaines why there is 33% less throughput on my 
algorithm than the Chriss Thomasson algorithm, but here is the big
news, my two locks algorithm that is a blocking algorithm is scoring
the same throughput as the Chriss Thomasson algorithm on the pop() side 
even though it uses more variables (4 variables) on the pop() side than 
the Chriss Thomasson algorithm that generate data movements between 
caches and between the memory system and the local caches,
why ? cause the in the two lock algorithm there is less contention
on the "Bus" inside the critical section, so the two lock algorithm
is more efficient cause it reduces efficiently the contention on the Bus.


Thank you,
Amine Moulay Ramdane.

[toc] | [next] | [standalone]

#2286

On 4/28/2014 12:35 AM, aminer wrote:
>
> Hello,
>
> Follow with me please...
>
>
> I have just explained to you why the following Chriss Thomasson
> algorithm:
>
> https://groups.google.com/d/topic/lock-free/acjQ3-89abE/discussion
>
> has scored 33% more throughput on the pop() side than my algorithm..
>
> Here is my algorithm:
>
> http://pages.videotron.com/aminer/CQueue2.htm
>
>
> So follow with me please cause we have to be smart to understand
> the inside of those scalable algorithms...
>
> I have just said that since i am using more variables that generate
> data movements between caches and between the memory subsystem and the
> caches, those data movements causes a contention on the "Bus" system
> and the Bus system must serialize those data movements, and more than
> that since there is contention in this scenario as i have just explained
> this "factor" that we call "contention" in this scenario
> is highing the waiting time of the threads and is lowering the

I mean is "highering" not highing.

> throughput,so since i am using 4 variables that generate data mouvements
> between caches and between the memory system and the caches
> and since the Chriss Thomasson is using 3 variables, so my algorithm
> is causing more contention on the Bus than the Chriss Thomasson
> algorithm and this explaines why there is 33% less throughput on my
> algorithm than the Chriss Thomasson algorithm, but here is the big
> news, my two locks algorithm that is a blocking algorithm is scoring
> the same throughput as the Chriss Thomasson algorithm on the pop() side
> even though it uses more variables (4 variables) on the pop() side than
> the Chriss Thomasson algorithm that generate data movements between
> caches and between the memory system and the local caches,
> why ? cause the in the two lock algorithm there is less contention
> on the "Bus" inside the critical section, so the two lock algorithm
> is more efficient cause it reduces efficiently the contention on the Bus.
>
>
> Thank you,
> Amine Moulay Ramdane.
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>

[toc] | [prev] | [standalone]

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

csiph-web