On Mon, 17 Oct 2005, Adrian Chadd wrote:
Hm, not if you checked every time the client side had written some data to the
client - if at this point the bucket has filled up a little then you
immediately delete the timed event and reschedule IO.
This would probably lead to heavy unfairness between the clients of the
pool I think..
I'll have more of a think about it after I've completed removing the
current commDefer stuff and its been tested. What do i have to do to get more
people to test - re-implement kqueue/epoll? :P
Probably.
For me looking into deferred reads is not much of a point without also
looking into handling very many fds reasonably efficient. Even the most
efficient poll set management implementation quickly spirals out of hand
when the number of fds grows..
Regards
Henrik
