U++ forum: Welcome to the forum

Re: Thoughts about alternative approach to multithreading [message #19189 is a reply to message #19188]

Wed, 19 November 2008 01:45

Mindtraveller
Messages: 917
Registered: August 2007
Location: Russia, Moscow rgn.

Experienced Contributor

Thank you for your attention to this topic.

* You don't do anything in case two callback functions have same hash.
--- Yes, and this is the problem for now. I`d like to use HashBase as hash table (it would solve this problem) but it is not documented and is rather a black box for me when I compare my plain realization and HashBase class efficiency.

* is "curArgsQueue.AddTail(newArgs);" (and the rest of them) atomic? Or the RequestAction can't be called concurrently for the same hash (function)?
* I was unable to copy/paste that piece of code and make it work. (should it work with 2008.1 + MINGW? Can you post some test package?)
--- Certainly, because I posted something like petrified version of the original code, just to discuss an idea itself.

* P1 and P2 must be moveable, right?
--- I keep thinking about this requirement. This is crucial for overall efficiency, but non-movable parameters should work also< I suppose...

* Also, I believe that all event queues will have to be synchronized anyway, something that your current measurements do not account for....
--- I mentioned before that actual tests were made for working class, not a simple & rather abstract example I`ve shown.

* My MT server (the application I have developed this year, it is the backend for community website) is able to handle 10000 requests per second (on quadcore CPU). Every such request would generate tens or even hundreds 'events' in your model.
--- I believe it is time to make some comparison. May be we should choose some reference example to be rewritten using alt. threading. It will give some comparisons between classic and alternative approach efficiency. Also, may be CoWork reference example is not the best choice to compare efficiency here? IMO it contains some functionality not covered by Thread alternative (if alt. approach will be proved to be worthy, I will make sense to make CoWork alternative, but not earlier). May be we should choose some comparison example wich is closer to "pure" threading? Or may be I`m wrong and CoWork reference example is just what we need to start comparison with? What do you think?

[Updated on: Wed, 19 November 2008 01:58]

Report message to a moderator