Intel® Moderncode for Parallel Architectures
Support for developing parallel programming applications on Intel® Architecture.

Parallel Hashlist was updated to version 1.43

aminer10
Novice
285 Views

Hello all,


Parallel Hashlist was updated to version 1.43


In the previous version i have set the number of lightweight
MREWs(multiple-readers-exclusive-writer)
to 128 but i have decided to change that in version 1.43 so that you can
give a variable number of MREWS, this will scale better and now the
constructor will look like this:

hash1:=TParallelHashList.create(trait, Hashnize, number_of_MREWS);

and the number_of_MREWS must be less or equal to the Hashsize

also i have given you two examples inside the zipfile, but please note that
the IntToStr() that i am using insode the test files don't scale well, but
in reality
and in fact parallel hashlist does scale very well if you don't use
IntToStr()..

Description:

A parallel HashList with O(1) best case and O(log(n)) worst case access that
uses lock striping and lightweight MREWs(multiple-readers-exclusive-writer)
,
this allows multiple threads to write and read concurently. also
parallelhashlist
maintains an independant counter , that counts the number of entries , for
each
segment of the hashtable and uses a lock for each counter, this is also for
better scalability.

Note: When i have done those benchmarks , there was not enough/much items
organized as a self-balancing tree in the individual chains of the
hashtable, so ,
almost all the items was found and inserted in O(1) , so the parallel part
in the
Amdahl equation was not much bigger compared to to the serial part. But you
will notice in pratice that as soon as you will have more items on the
chains of
the Hash table , organized as self-balancing tree, with a worst case log(n)
, the
parallel part will become bigger in the Amdahl equation and you will have
better
performance and scalability than the numbers in the graph of the benchmarks
...
Please pass a hashsize and the number of mrews in power of 2 to the
constructor
by using the shl operation for example like this

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

Why do you have to use a power of 2 ?

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 using modulo functions inside parallelhashlist, and using a number of
locks in power of 2,
so you have to use hashsize in power of 2 , this will make the modulo
function of the delphi
and freepascal compilers 10X faster.


You can download parallel hashlist from:


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



Sincerely,
Amine Moulay Ramdane.


0 Kudos
1 Reply
aminer10
Novice
285 Views

Hello,
I have updated parallelhashlist to version 1.44
What have changed in version 1.44 ?
I have corrected a bug in the constructor,
Before, itwas:

if (AHashSize mod size2) <> 0 then size2:=size2+1;
In version 1.44 i have corrected the bug by changing size2 to size1:
if (AHashSize mod size1) <> 0 then size2:=size2+1;
And after that i am setting correctly the array fcount1(the independant counters ,
that count the number of entries , for each segment of the hashtable) and the
array mrew((multiple-readers-exclusive-writer locks) like this:
setlength(fcount1,size2);
for i:=0 to size2-1 do fcount1:=0;
for i:=0 to size2-1 do mrew:=TOmniMREW.create;

You can download parallel hashlist from:


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

Sincerely,
Amine Moulay Ramdane
0 Kudos
Reply