begin quoting Tracy R Reed as of Sat, Dec 29, 2007 at 01:28:25PM -0800: > [EMAIL PROTECTED] wrote: > >Since web servers cannot just push updates/data to client browsers with > >HTML, > >the poor client must often keep *polling* the server in web land. > >This isn't scalable for obvious reasons. > > Look at how SIP works. You subscribe to a resource and when something > interesting happens to that resource (phone call, voicemail, SMS > message, etc) you get a notification. But once again firewalls are going > to cause a problem unless...
...they get properly configured. If you're not going to configure a firewall, you've established that you don't WANT those sort of connections. Of course, now you have to build smarter firewalls, because the developers are busy figuring out ways around simple firewalls. Stateful traffic inspection, profiling, and fingerprinting seem to be the new tools. :-/ > >A clever solution is to replace polling with a single query that keeps > >sending whitespace to keep it alive. Rather than building and > >tearing down HTTP connections to server all the time (with polling), > >we can just work around the browser's timeout values by keeping one > >connection alive until server has data for it!!! > > ..we just keep the connection open all the time. That's a kludge. It > wastes a lot of bandwidth and requires the other end to keep file > descriptors open. And are you implying that whitespace takes up less > bandwidth than some other form of data? :) File descriptors are at a different level of abstraction. Keeping a connection open only forces a file descriptor to be open because that's the way we're used to using a TCP connection. Keep the file descriptor around, and the connection stays around; close the file descriptor, and the connection gets terminated. The network doesn't enforce this. As for bandwidth... well, sometimes it's instructive to watch tcpdump on a supposedly idle network. . . there's a lot of overhead out there. And until we have characterized the intervals between updates, it's all pretty useless anyway. If data shows up once a day, then polling works great; if data arrival is random but bursty, a combination of polling and keeping the connection open would probably work. Etc. etc. > >This seems to solve a lot of problems of HTML and have lots of potential > >for future dynamic AJAXy type web apps. > > I don't see anything new here. Indeed. -- See the new tricks, same as the old tricks, only not to solid, but with glitter! Stewart Stremler -- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
