Eric Eilebrecht's blog - All Comments

re: Windows I/O threads vs. managed I/O threads

DnshPly9 — Fri, 19 Oct 2012 06:31:31 GMT

I was looking for to use IO Thread from ThreadPool in C#. And your answer is what I was looking for.

Very nice blog. Many Thanks :)

re: CLR 4.0 ThreadPool Improvements: Part 1

Kapil — Sat, 18 Feb 2012 12:14:49 GMT

I'm trying to process a very high number of messages per second that arrive through a socket. As each message is received, I need to process it. The processing can be in any order irrespective of the receiving conditions. We are currently using the system.collections.generic.queue class and and we are creating threads using system.threading.thread class. The threads are continuous and we are using thread.suspend when a queue is empty and thread.resume when a new item is inserted in the queue. The time of processing is very important in our application. We are currently using CLR 3.5, Please tell us how to improve upon the performance.

re: Fun programming problem: a simple lock-free algorithm

Tanmay Chakrabarty — Sat, 03 Sep 2011 04:04:51 GMT

Sorry sorry sorry.....ho..ho...I did not missed the declaration of 'globals', what I missed is that in S5-S8 you declared end, a, b, start for the ReaderThread() function and not using those of the struct....

Thanks for replying.

re: Fun programming problem: a simple lock-free algorithm

Eric Eilebrecht - MSFT — Fri, 02 Sep 2011 22:34:56 GMT

Tanmay, immediately after the declaration of the struct, there is a declaration for a single global instance of that struct named "globals" (all lower case).

re: Fun programming problem: a simple lock-free algorithm

Tanmay Chakrabarty — Fri, 02 Sep 2011 14:53:55 GMT

I am a rookie and I did not understand how you used these end, a, b, start in S5,S6,S7,S8 ? I mean you haven't declared them as variables, I don't know the use of struct, but class view, an object is required to use those variables.

Off-topic: How will I know if you have replied to this comment? However, leaving my email address cse.tanmay@gmail.com

re: When does it make sense to use Win32 Fibers?

Eric Eilebrecht - MSFT — Tue, 19 Oct 2010 20:36:27 GMT

Ian: I don't know all of the history here, but there are certainly other reasons to want a public fiber API. It's not that fibers are only useful to SQL Server, it's just that they're not often useful to very many applications. That describes quite a few public Win32 APIs, actually. :)

re: When does it make sense to use Win32 Fibers?

Ian Ringrose — Mon, 18 Oct 2010 18:10:55 GMT

Is the fiber API only a public API to stop any-trust problem from other database vendors claiming that the OS team is helping the Microsoft SQL Server team to much?

re: CLR 4.0 ThreadPool Improvements: Part 2 (sort of...)

dimkaz — Wed, 15 Sep 2010 17:42:37 GMT

I guess it's not a big deal, In my testing (4 cores) I am loosing 2-3%.

Seems like specifying MaxDegreeOfParallelism is even a smaller deal to gain something :)

re: CLR 4.0 ThreadPool Improvements: Part 2 (sort of...)

Eric Eilebrecht - MSFT — Wed, 08 Sep 2010 17:04:23 GMT

In our testing, we do not typically observe a very large performance problem due to this probing. If you have a scenario that demonstrates this, I'd be interested in hearing more. What kind of processing are you doing? How big of an impact do you observe due to the extra threads?

re: CLR 4.0 ThreadPool Improvements: Part 2 (sort of...)

Dmitry Zaslavsky — Tue, 07 Sep 2010 21:29:02 GMT

Hi Eric,

It seems for the pure CPU bound code it's ALWAYS good to set MaxDegreeOfParallelism = cpuCount;

Otherwise threadpool wiil try to schedule more threads, trying to "see" if it will help.

I actually verified this in practice. It's not a huge win when the loop body is small (timewise) but it's huge with the body is expensive.

Just didn't see this recommendation, and wanted to see if you would agree