Joe Duffy and Igor Ostrovsky: Parallel LINQ under the hood

Posted By: Charles | Mar 18th @ 10:32 AM

Continuing our exploration of the Parallel Computing Platform and the folks who think it up and build it, we sit down with software developers Joe Duffy (he wrote the original Parallel LINQ Think Week paper that Bill Gates read and obviously enjoyed!) and Igor Ostrovsky (he spends his time implementing PLINQ and associated technologies) to deeply dig into how PLINQ works and what makes it tick from a parallel processing point of view.

Lots of whiteboarding and great conversation in this one, as is usual with the people helping to shape our general purpose parallel computing development platform.

Enjoy.

Low res file here.

Tags: Parallel Computing, Parallel Extensions, PLINQ

Rating:

Views (14,516)

Comments (5)

Downloads

gaurav.net

Awesome..
Dwnldin999....

staceyw
Before C# there was darkness...

Great. Thanks again Joe, Igor, and Charles.
Forcing tasks (i.e. Charles question) on many cores explicitly actually has a simple use case. Correctness debugging. If my code is tricky in terms of the delegate bodies, I may want to ensure all tasks perform on different Procs, even if the library feels 1 thread is enouph. Because your race bugs will not show up using 1 proc or even 1 thread passed around on many procs.

ymasuda_
ymasuda_

Despite difficult theme of parallel algorithm computing, this talk delivered developers friendly knowledge of parallel data management. Supercomputing database language parsers and engines may have this I/O capability, and .NET languages are bound to constitute parallel processing algorithms. For developers business investment, given Parallel language and processors should have to expose extensibility points where .NET developers consider .NET Remoting and Web Services as an area of distributed computing and extensible execution stack to include parallel.

Likewise Windows kernel level thread context management, PLINQ may supply fine granular data access management for managed Threads, on the other hand Windows Clustering and Virtualization may take up supervisor on clustering nodes, and Web Farm and Data Grid may do load balancer similarly.

Development partners should be thankful for fine logical explanation of new ideas PLINQ.

littleguru
allein, allein,... allein, allein!

Thanks for this video guys! Smiley

sokhaty

Good video. Too short apparently to cover the subject deep enough Smiley

General suggestion to the PLINQ team - please do provide programmers with extra knobs and switches to fine tune parallel execution flow.
Based on personal experience with SQL optimizers including the parallel ones, out of five more or less complex queries one would need a hint to run correctly.
I wouldn't expect PLINQ be any different in this regard, especially considering that it will work on data that won't have any statistics associated with them.