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

Groups > comp.programming.threads > #2270 > unrolled thread

Amine, you have said this

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

Contents

  Amine, you have said this aminer <aminer@toto.net> - 2014-04-26 21:46 -0700

#2270 — Amine, you have said this

Question:

Amine, you have said this:

 > So this why the Chriss Thomasson algorithm has scored more throughput
 > on the pop() side , cause on my pop() method there is more variables
 > that generate data movements between caches and can cause contention.


How can you say this and you know that his algorithm has the same number 
of variables than your algorithm, so where is the problem ?


Answer:

I have to be more clearer and precise, what i wanted to say is this:

His "@c^.ver" variable do not generate lots of contention, cause
when the poping threads will pop one after the other in parallel
they will touch less number of time "@c^.ver" than my algorithm,
when the thread will enter the pop() method in the Chriss Thomasson 
algorithm it will incur one data movement and cache-line transfer
on "@c^.ver" on the pop() side,  and that's optimal and efficient , but 
in my algorithm i am using a Ticket mechanism that incur more data 
movements between caches and this also cause contention on the "Bus" 
system, although i am using a backoff mechanism to reduce the contention 
on the "Bus" system, there is still more contention on my algorithm than 
the Chriss THomasson algorithm, cause my "tail" variable on the pop() 
side generate more data movements between caches than
the Chriss Thomasson concurrent FIFO queue, and when we say
more data movement between caches this means also more contention
on the Bus system and this means also more waiting time...
this is why my algorithm has scored 33% less throughput than
the Chriss Thomasson concurrent FIFO queue that uses the bakery algorithm.



Thank you,
Amine Moulay Ramdane.

[toc] | [standalone]

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

csiph-web