topic Parallelhashlist was updated to version 1.41 ... in Intel® Moderncode for Parallel Architectures

Parallelhashlist was updated to version 1.41 ...

aminer10 — Wed, 30 May 2012 22:33:46 GMT

Hello all,

I have updated parallelhashlist to version 1.41..

I have modified parallelhashlist to use 128 locks, but you
have to use a hashsize inpower of 2 and greater to 1000,
how do you pass a hashsize in power of 2 to the constructor?
by using shl operation like this

trait:=TCaseinsensitiveTraits.create;;
hash1:=TParallelHashList.create(trait,1 shl 25);

Why do you have to use a power of2 ?

Please read this:

Power-of-Two Hash Table Size

"Any data structures 101 book will say choose a prime for the
number of buckets, so that the bucket's index can easily be computed
by h(k) = k mod m, where k is the key value and m is the bucket size.
While this approach is straight-forward, there are a number of issues
with it, including slow modulo performance. ConcurrentHashMap instead
uses a power-of-two rule"

http://work.tinou.com/2008/09/performance-optimization-in-concurrenthashmap.html

I am usingmodulo functions inside parallelhashlist, and using a number of locks
in power of 2, so you have to usehashsize in power of 2 , this will make the
modulo function ofthe delphi and freepascal compilers10X faster.

You can download parallelhashlist version 1.41 from:

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

Thank you.

Amine Moulay Ramdane.

Parallelhashlist was updated to version 1.41 ...

aminer10 — Wed, 30 May 2012 22:38:46 GMT

I wrote:
>I have modified parallelhashlist to use 128 locks

I mean 128 lightweight MREWs (multiple-readers -exclusive-writer).

Thank you.

Amine Moulay Ramdane.

Parallelhashlist was updated to version 1.41 ...

aminer10 — Wed, 30 May 2012 22:46:07 GMT

I wrote:

>I am usingmodulo functions inside parallelhashlist, and using a number of locks
>in power of 2, so you have to usehashsize in power of 2 , this will make the
>modulo function ofthe delphi and freepascal compilers10X faster.

I mean that the modulo function that i am using a lot inside parallelhashlist
is 10X faster when the hashsize is power of 2 than when it's not power of 2.

Thank you.

Amine Moulay Ramdane.

Parallelhashlist was updated to version 1.41 ...

aminer10 — Sat, 02 Jun 2012 22:40:08 GMT

Hello,

I have done some scalability tests on parallelhashlist 1.41
on an Intel Core 2 Quad Q6600..

Using 4 threads on4 coresusing the find() method i had a perfect scaling of 4X .

Using4 threads on4 coresusing the Add() methodihad a scaling of 3.4X

Regards,
Amine Moulay Ramdane.

Parallelhashlist was updated to version 1.41 ...

aminer10 — Sat, 02 Jun 2012 22:52:29 GMT

Hello,

I have done those tests to show some numbers...

But the scalability of parallelhashlist does depend onthe data keys
that are presentedto parallelhashlist,when the data keys are hashed
and are very well distributedacrossthe segments, in fact you can
get a perfect 4X scaling both on reads and writes.

Thank you.

Amine Moulay Ramdane.

Parallelhashlist was updated to version 1.41 ...

aminer10 — Sat, 02 Jun 2012 22:58:04 GMT

I write:
>But the scalability of parallelhashlist does depend on the data keys
>that are presented to parallelhashlist, when the data keys are hashed
>and are very well distributed across the segments, in fact you can
>get a perfect 4X scaling both on reads and writes.

I mean a perfect scaling on 4 cores...

Regards,

Amine Moulay Ramdane.