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

Groups > comp.programming.threads > #2357 > unrolled thread

About my new algorithm

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

Contents

  About my new algorithm aminer <aminer@toto.net> - 2014-05-18 15:10 -0700
    Re: About my new algorithm aminer <aminer@toto.net> - 2014-05-18 15:12 -0700
    Re: About my new algorithm aminer <aminer@toto.net> - 2014-05-18 15:16 -0700

#2357 — About my new algorithm

Hello,


Here is what have changed in my new algorithm of a very fast concurrent 
FIFO queue: In my previous algorithm the atomic increment of 
"LockedIncLong(temp)" was located after the getlength() method , so in 
the worst case scenario,  that happen frequently under contention, we 
have to transfer 4 cache-lines between the cores: the tail variable, the 
head variable, and another time the head variable,  and the flag1 
variable, this will make it 4 cache-line transfers and this is 
expensive... so in my new algorithm i have reduce that to 3 cache-lines 
transfers in the worst case, that happen
frequently under contention, and this is much better , so my new 
algorithm has scored much better throughput than my previous algorithm,
and my new algorithm is 100% starvation-free.

Hope you will be happy with this new algorithm.

Here is my new algorithm:

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


And you can download my very concurrent FIFO queue version 1.2 from:

https://sites.google.com/site/aminer68/concurrent-fifo-queue-1



Thank you,
Amine Moulay Ramdane.

[toc] | [next] | [standalone]

#2358

Hello,

I was speaking about the push() method.



Thank you,
Amine Moulay Ramdane.

[toc] | [prev] | [next] | [standalone]

#2359

I correct, please read again...


Hello,


Here is what have changed in my new algorithm of a very fast concurrent 
FIFO queue: inside the push method of my previous algorithm, the atomic 
increment of "LockedIncLong(head)" was located after the getlength() 
method , so in the worst case scenario,  that happen frequently under 
contention, we have to transfer 4 cache-lines between the cores: the 
tail variable, the head variable, and another time the head variable, 
and the flag1 variable, this will make it 4 cache-line transfers and 
this is expensive... so in my new algorithm i have reduced that to 3 
cache-line transfers in the worst case, that happen
frequently under contention, and this is much better , so my new 
algorithm has scored much better throughput than my previous algorithm,
and my new algorithm is 100% starvation-free.

Hope you will be happy with this new algorithm.

Here is my new algorithm:

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


And you can download my very concurrent FIFO queue version 1.2 from:

https://sites.google.com/site/aminer68/concurrent-fifo-queue-1



Thank you,
Amine Moulay Ramdane.

[toc] | [prev] | [standalone]

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

csiph-web