Back to Questions
2.4k
views1
comment
Parallel tests - multiple cores - wired results
f#parallel

Hi,

How well does F# scale with a machine with multiple cores?

I played with the code below and I got wired results.

1
2
3
4
5
6
7
8
9
10
11

Machine 1 

Time: 05min 27s

Cores: 16

Machine 2

Cores: 48 Cores

Time: 06min 15s

During the test all cores on both machines got used as expected.

For this small code the difference is small, on my real app the diference is more than 10min.

There is any config that might help?

1
2
3
4
5
6
7
8
9
10
11
12
13
14

let isPrime n =

    let rec check i =
        i > n/2 || (n % i <> 0 && check (i + 1))
    check 2

let nums = [|1..5000000|]

let finalDigitOfPrimes = 
        nums 
        |> PSeq.filter isPrime
        |> PSeq.groupBy (fun i -> i % 10)
        |> PSeq.map (fun (k, vs) -> (k, Seq.length vs))
        |> PSeq.toArray

Tkz

JRG

.

Hi,

I don't know why exactly performance degrades with more cores. But PSeq is simply a thin layer around parallel Linq, and I believe PLinq by default uses partitioning strategies which are probably not optimal for this problem.

Fortunately it is possible to customize this. Using a load-balancing partioner clearly improves performance even on my lowely 2-core machine. Example code:

1
2
3
4
5
6
7
8

 

let finalDigitOfPrimes = 
   System.Collections.Concurrent.Partitioner.Create(nums, loadBalance=true).AsParallel() 
   |> PSeq.filter isPrime 
   |> PSeq.groupBy (fun i -> i % 10) 
   |> PSeq.map (fun (k, vs) -> (k, Seq.length vs)) 
   |> PSeq.toArray

Would be interesting to know how this performs on a 48-core machine...

By on 4/21/2011 12:35 PM ()

    Topic tags

    IntelliFactory Offices Copyright (c) 2011-2012 IntelliFactory. All rights reserved.
    Home | Products | Consulting | Trainings | Blogs | Jobs | Contact Us | Terms of Use | Privacy Policy | Cookie Policy     		Built with WebSharper