On Sun, 2006-08-27 at 10:49 +0200, Roland Weber wrote: > Hi Oleg, > > > I have the first version of an ultra trivial asynchronous I/O event > > dispatcher ready. My implementation is largely based on the I/O Reactor > > pattern of Doug Lea. The I/O dispatcher does not do much at the moment > > but I already have a runnable version of an echo server based on it. > > Surprisingly, that takes very little code. > > So I take it that NIO is advantageous primarily on the server side. > That solves a few of the tihings that have puzzled me so far about > HttpNIO. >
Hi Roland, NIO is just not suited for protocols like HTTP. Period. The performance overhead of NIO is just too great for a moderate number of short-lived connections where data comes in tight bursts. NIO starts paying off only when the number of simultaneous connections exceeds a thousand and most of the time those connections stay idle. At this point of time I tend to think NIO on the client side makes no sense at all, as it makes little sense for an HTTP client to keep hundreds of connections open. > > Is anyone interested in seeing the intermediate results of my work and > > giving my an early feedback? Shall I start committing code early, or > > shall I hold off until the whole HTTP transport is more or less ready > > and the API is more or less feature complete? > > I won't be able to give you feedback on NIO, since I don't have the > time or inclination to learn enough about NIO in the near future. No need to review the NIO code. Just take a look at the public API and let me know if it "feels" okay. > Regarding source repositories, I usually check in as soon as stuff > compiles. In this case I'm wary because of the pending changes with > the connections. I do not expect any changes in the HttpServerConnection. It is the client side of the story that needs work. > On the other hand, code will have to be adapted > whether it's in the repository or not. So go ahead an commit the > new stuff if you want to. > I'll start checking in classes, which I have a certain degree of confidence in not being completely wrong. Cheers, Oleg > cheers, > Roland > > --------------------------------------------------------------------- > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
