Just wanted to capture the idea of not having receive() do a notify until the connection is 1/2 full.
I did this experimentally in my C++FBP implementation, and I noticed that it improved performance significantly. The down side is that you may have to add some flush logic to the connection to tell whether the downstream process can be reactivated, and I think in the deadlock logic also...
John C. suggested doing it on the send side as well, but I'm not sure how to do that...
I have to think about this and study some more code to understand what is going on. At the moment I focus on Network, cause I messed up the deadlock code and I need to revert it to its original state (yes, I f****d up).
This does beg the question: how are we going to keep track of our wishes, thoughts and ideas, and at the same time keep focussed on the highlights?
Does sourceforge have a bug-mechanism? Could we use it for features. so we have a list and can prioritise things (if that is english at all)?
Log in to post a comment.