Most Common Performance Issues in Parallel Programs

IHateSpaghetti {code} — Tue, 12 Aug 2008 10:56:59 GMT

The Parallel Programming with .NET (pfxteam) has published a list of issues why a concurrent program

re: Most Common Performance Issues in Parallel Programs

Omer van Kloeten — Wed, 13 Aug 2008 08:30:59 GMT

Will the issue with GC be mitigated by using Concurrent GC or is Server GC the only (best?) solution?

Reflective Perspective - Chris Alcock » The Morning Brew #157

Reflective Perspective - Chris Alcock » The Morning Brew #157 — Wed, 13 Aug 2008 10:18:55 GMT

PingBack from http://blog.cwa.me.uk/2008/08/13/the-morning-brew-157/

Most Common Performance Issues in Parallel Programs | Igor Ostrovsky Blogging

Most Common Performance Issues in Parallel Programs | Igor Ostrovsky Blogging — Wed, 13 Aug 2008 11:15:38 GMT

PingBack from http://igoro.com/archive/most-common-performance-issues-in-parallel-programs/

Most Common Performance Issues in Parallel Programs

MS Consulting Blog — Wed, 13 Aug 2008 15:37:44 GMT

Most Common Performance Issues in Parallel Programs

re: Most Common Performance Issues in Parallel Programs

David — Fri, 22 Aug 2008 19:15:10 GMT

"In such cases, though, the asynchronous programming model would likely result in even better performance."

So, we still need APM?

re: Most Common Performance Issues in Parallel Programs

toub — Sat, 23 Aug 2008 23:25:23 GMT

Omer, the concurrent GC doesn't really mitigate the issue like the server GC does.

David, yes. But we're investigating making it easier to use the Task Parallel Library to help use the APM. It's already possible to create Task/Future<T> wrappers around Begin/End calls, such as with the example at http://blogs.msdn.com/pfxteam/archive/2008/03/16/8272833.aspx.

re: Most Common Performance Issues in Parallel Programs

Frank Bakker — Wed, 03 Sep 2008 12:01:34 GMT

In my current project I have a scenario where I loop over 10.000 to a 100.000 items. For each item I read some data from a database (IO), do some conversion logic (CPU) and write the resulting XML to disk (IO). The total time for processing of a single item is about 50 ms. The processing of each item is independant from the processing of every other item, so this could very well be done in parallel and would be a good candiate for the TPL.

As I understand it, the TPL's scheduler is optimised for scenario's with CPU intensive tasks. I wonder how it will perform in this scenario where tasks are both CPU and IO intensive, will the CPU idl while a task is waiting for IO, or is the next task scheduled in to take over?

My experience with asyncrounous IO is that the thread hopping can quickly obfuscate the code, so I would reather not go there.

re: Most Common Performance Issues in Parallel Programs

igoro — Fri, 05 Sep 2008 02:25:06 GMT

Frank,

We are still making changes to the TPL scheduler, so I can't say for sure how precisely it will interact with I/O.

But, you are right in saying that TPL is optimized for execution of CPU-intensive tasks. Even if the tasks are I/O bound, you may still see a speedup if the I/O device can service all the requests coming from multiple CPUs. That may be the case for example if you are reading from a disk array, or sending requests over the network to multiple servers.

On the other hand, if much of the time is spent reading from a single harddisk, then concurrency probably won't make things any faster.

Hope this helps explain things.

Intel Software Network Blogs » Eliminate False Sharing? Wrong!

Intel Software Network Blogs » Eliminate False Sharing? Wrong! — Fri, 10 Oct 2008 04:30:16 GMT

PingBack from http://software.intel.com/en-us/blogs/2008/10/09/eliminate-false-sharing-wrong/

re: Most Common Performance Issues in Parallel Programs

john.rayner — Wed, 26 Nov 2008 02:03:36 GMT

How do you detect that false cache-line sharing is an issue for an application? Are there any tools that can monitor requests to main memory in order to detect this?

re: Most Common Performance Issues in Parallel Programs

igoro — Wed, 26 Nov 2008 02:47:45 GMT

John,

This MSDN Magazine article discusses false sharing in more detail, including tools that can be used to detect it: http://msdn.microsoft.com/en-us/magazine/cc872851.aspx

Igor

Parallel Programming with NET Most Common Performance Issues in | work from home

Parallel Programming with NET Most Common Performance Issues in | work from home — Tue, 16 Jun 2009 14:55:48 GMT

PingBack from http://workfromhomecareer.info/story.php?id=22777