On Jul 24, 2006, at 4:15 AM, [EMAIL PROTECTED] wrote: Looking at "To thread or not to thread", processing the reads on a separate thread from the selectors is not something I have tried. Has anyone else tried that with MINA? My gut instinct would be that it might help with a very large number of clients but for smaller numbers of connections (e.g. hundreds) the extra context switch will not help? I haven't tried that, but boosting the number of IoProcessors to scale with the number of CPU cores did help. In AMQP, we are using the idle mechanism in MINA to implement heartbeats. If we miss enough heartbeats, we close the channel which does cancel the SelectionKey. Is that not enough ensure that selection keys set gets cleaned up? Even if I were implementing http I would use the idle mechanism to kill inactive connections. Am I missing the point? That accomplishes the same end-goal.. But not everyone that does NIO may implement the idle mechanism that MINA has :) -pete |
smime.p7s
Description: S/MIME cryptographic signature
