- From: David - Morris <dwm@shell.portal.com>
- Date: Sat, 17 Dec 1994 19:22:38 -0800 (PST)
- Cc: http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
On Sat, 17 Dec 1994, Ari Luotonen wrote: > If I'm not totally lost in space, keep-connection will not even work > with all TCP implementations. There is no way to know beforehand that Well, I must admit I've thought about the problem rather than having had to implement, but if the TCP/IP connection can't be kept open then all kinds of things are eventually going to break .... like simple TELNET connections into that host. The CGI scripts should be fixed if necessary to avoid behaviors which close the connection. It could also be possible for a server to be paramatized to refuse keepalive when certain child processes are invoked. The headers sent with each response should confirm the intent to continue keeping the connection open so the server can tell the client when to expect the current transmission to end the connection. There is nothing wrong with MGET that would precluding implementing it that I know of but it will require more code changes in clients and servers than simply keeping a connection open. The connection open can be tuned to serve many documents if desired and for a proxy/server connection can remain open over a very long interval if desired potentially handling many requests. (There are security issues but then caching satisfied by a proxy has the same problem since cache serving is logically a long proxy/server connection.) To solve the oldproxy new server problem, a new hack in the server would provide a workaround by allowing the admin to specify proxys who can't handle open connections. The reply would go back as today. But as has been suggested some real experience to back up the simulations Simon and Jeff have reported would be helpful. Dave Morris
Received on Saturday, 17 December 1994 19:23:39 UTC