Talk:Concurrent computing

Random order?

Is random order what is meant here, or is "any order" sufficient? --Short Circuit 00:16, 6 February 2007 (EST)

Random order means that the order of output is not determined at compile time. The order of output may differ from one execution of the program to another due to differences in timing between the concurrent units providing the output. --Waldorf 11:47, 7 February 2007 (MST)
"Undetermined" might be the more appropriate word. At least one of the programming examples attempts to enforce random behavior. --Short Circuit 14:17, 7 February 2007 (EST)
Random is what I wanted. The Ada example I provided creates three separate tasks. Each one calculates a random number between 0.0 and 1.0. The task then delays the number of seconds corresponding to that random number. A delay of 0.5 lasts 0.5 seconds. There are two goals to "Simple concurrent actions". Those goals are to demonstrate the syntax for defining concurrent behavior within a single program, and the syntax for creating a random sleep or delay. --Waldorf 17:59, 7 February 2007 (MST)
I see. You might want to explain that in the task; I thought the task's purpose was simply to demonstrate forking. --Short Circuit 20:33, 7 February 2007 (EST)
What do you need to be clarified? I specified threads, tasks or co-routines. It can be argued that forking is a form of concurrency, which is why I specified threads, etc. Please let me know how you would like the task description to be clearer. There is currently a separate task to demonstrate forking. I have plans to explore more aspects of concurrency with tasks demonstrating synchronous and asynchonous communication between threads, etc. --Waldorf 19:27, 7 February, 2007 (MST)
OK. I think I understand now. Your task description is probably fine. --Short Circuit 08:19, 8 February 2007 (EST)
AFAIK Co-routines are deterministic because there is only one thread of execution at any point of time. (deterministic output). Rahul 14:48, 9 December 2008 (UTC)

Race conditions?

It seems to me that none of the examples so far make sure that there's no attempt to output two strings at the same time. Do all the languages have reentrant output routines, or are there examples with subtle bugs in here? --Ce 09:33, 28 February 2008 (MST)

I'm 99% sure Java will not output two strings at once. The JVM handles it. I don't know enough about the other languages to answer for them. --Mwn3d 09:43, 28 February 2008 (MST)
Interesting. I believe it is not an issue since writing to a stream should be atomically done, or done in a thread-safe way. The worst that can happen is messing up output, but race conditions or deadlocks or any other bad thing should not occur (at least, if messing up output is not a bad thing! EnRosejotCotayde?!) Consider that the same stream (stdout) can be transparently used by several processes (e.g. I am running Kate editor from the same shell I use to compile code, so sometimes stderr of Kate is intermixed with stdout and stderr of other tasks) --ShinTakezou 18:50, 17 December 2008 (UTC)
Hm, maybe I am wrong, since for different processes the file descriptor is replicated; nonetheless they are always attached to the same terminal, and I am still thinking that race conditions are not possible. --ShinTakezou 19:01, 17 December 2008 (UTC)

What are "concurrent threads"?

Conceptually speaking, unless you have multiple CPUs, there can be only one "thread" that is executing at any one time. This suggests a possible approach for implementing this task in languages which do not "support threading": One could implement a simplistic thread scheduler, and then use that to implement the task.

Consider also languages which support concurrent syntax without necessarily haven gotten around to implementing support for dispatching across multiple CPUs.

This leaves me wondering, what does this task specification really mean, in a language agnostic sense? --Rdm 15:38, 24 November 2010 (UTC)

"in a language agnostic sense" ... I don't know. I'd muse that it mostly means, "of minimally-synchronized execution contexts", but then we'd have to come up with a meaning for the constraint 'minimal'. --Michael Mol 18:40, 24 November 2010 (UTC)
If the language/runtime supports executing over multiple CPUs, that's what should be done for this task. Otherwise, if the language/runtime conceptually supports concurrency but doesn't actually do so, it should be noted in the text for that solution that this is the case (probably along with a note as to what version of the lang/rtlib this is referring to, so that this is known to be an issue at that particular point in time instead of something that is an issue for all eternity). –Donal Fellows 16:42, 25 November 2010 (UTC)
Ok, I have taken a stab at this... --Rdm 14:22, 26 November 2010 (UTC)
Return to "Concurrent computing" page.