PLINQ performance problem

Question

I using PLINQ on collection of more then 10000 element... i m not getting the performance as compare with sequential query.

my system configuration is as follows: OS - Windows 7 32-bit, processor - Intel Core2Duo.

Please help me i am not getting the proper performance.

The query is:

ParallelQuery<int> j = Enumerable.Range(0, 1000000).AsParallel();
var sss =  j.Where(o => o%2 == 0);

You'll need to post the query or at least as much of it as possible to be able to get an answer. — Eamon Nerbonne, Jan 29 '11 at 11:12
ParallelQuery j = Enumerable.Range(0, 1000000).AsParallel(); var sss = j.Where(o => o%2 == 0); — slash shogdhe, Jan 29 '11 at 11:20
well i dont know y u have given me the down points as this is not realted to code alot — slash shogdhe, Jan 29 '11 at 11:21
@slash shogdhe: You can **EDIT** your question by clicking the edit link below, instead of posting some comments. — Cheng Chen, Jan 29 '11 at 11:25
btw, I've faced the problem that looks like yours, looking forward for updated question. — Restuta, Jan 29 '11 at 11:26
@restuta : well this is really a great problem,if i look onto the cpu status,my both processor start working if i run this query but not getting the performance — slash shogdhe, Jan 29 '11 at 11:31
I'd guess the cost of spliting it to multiple threads and rejoining the results is larger than the cost of the where. You might only get a speedup if your predicate is expensive. — CodesInChaos, Jan 29 '11 at 11:35
Hm, it looks strange, cos I can manually split and join and will get perf. gain. — Restuta, Jan 29 '11 at 11:38
@codeinChoas: yeah this might be possible,i will look onto it will create a expensive predicate and test it — slash shogdhe, Jan 29 '11 at 11:38

score 4 · Answer 1 · edited May 23 '17 at 11:47

As your tasks are very light-weight, a major overhead will be caused by thread context switching. Things can be improved by grouping the light-weight tasks into batches that will involve more substantial amounts of work. This can be achieved using partitioning support in PLinq. Please see the following question:

Parallel Operation Batching

By grouping the tasks you will make better use of multiple threads as the management overhead becomes less significant.

In your existing code example it is akin to fetching water from the well with a thimble, grouping the tasks is more like using a bucket.

score 2 · Answer 2 · answered Jan 29 '11 at 12:09

Even though the task is split up among threads the overhead of invoking the threads can all too often easily become more than simply running it sequential.

In this case you are doing some really simple math on it, so each operation is in the <10nsec and utilizing L1 and L2 cache very well. When you introduce threads they have a relatively high startup cost, partially because they are sleeping while waiting for someting to do.

Have a look at Custom Partitioners for PLINQ and TPL.

PLINQ performance problem

2 Answers2