tag:blogger.com,1999:blog-35024958.post2234533118841563732..comments2011-04-05T19:45:29.606+10:00chrishttp://www.blogger.com/profile/09159109667366328919noreply@blogger.comBlogger4125tag:blogger.com,1999:blog-35024958.post-16312707810585563232011-04-05T19:45:29.606+10:002011-04-05T19:45:29.606+10:00<b>Awesome</b>.<br />It remains to get rid of boost::function. boost::function can add additional overhead related to memory allocation.<br />What about overall custom memory allocation support? <br />What about replacing boost::function? For example, it could be replaced with boost::function with custom allocator forwarding to asio custom memory allocation.<br />Support of custom memory allocation could provide an opportunity to compete (sometimes) with Intel TBB (with TBB&#39;s task class).<br />I am more inclined to think that, thanks to its versatility and generalizations, asio can become the basis for the fork/join library, which is then able to enter the Boost.Marathttp://www.blogger.com/profile/01714473137005987457noreply@blogger.comtag:blogger.com,1999:blog-35024958.post-77432655847099287802008-10-13T17:36:00.000+11:002008-10-13T17:36:00.000+11:00Nice post Chris,<BR/><BR/>It would be nice seeing the recently-accepted Phoenix library help out those stuck with compilers that don't support C++0x lambda's.<BR/><BR/>Just thinking aloud: what is the overhead of using an io_service instance to run arbitrary function handlers? I've been doing this to implement Active Objects, but until now I'm not entirely convinced that using an io_service instance per active object is a "lean" enough way of going about it.<BR/><BR/>I hope you can write about that issue, as well as how to be able to define a custom error/exception handler for function objects scheduled for execution in an io_service queue.<BR/><BR/>Keep up the great work!Dean Michaelhttp://www.blogger.com/profile/12475688728121462783noreply@blogger.comtag:blogger.com,1999:blog-35024958.post-1208365887158659642008-10-11T22:35:00.000+11:002008-10-11T22:35:00.000+11:00inspiring post!<BR/><BR/>have you compared the performance gain vs a thread pool or map-reduce model ?<BR/><BR/>It would be nice to have a map-reduce example using asio. I guess it would be much cleaner than the coinvoke code !<BR/><BR/>joseAnonymousnoreply@blogger.comtag:blogger.com,1999:blog-35024958.post-28457027780267801782008-10-11T15:04:00.000+11:002008-10-11T15:04:00.000+11:00awesome! - one question though, could there be an overload of coinvoke that takes an iterator range of tasks?Arash Partownoreply@blogger.com