Path: utzoo!mnetor!uunet!lll-winken!lll-tis!ames!pasteur!ucbvax!SPAM.ISTC.SRI.COM!robert From: robert@SPAM.ISTC.SRI.COM (Robert Allen) Newsgroups: comp.protocols.tcp-ip Subject: 4.3 TCP sockets revisited. Message-ID: <8804192115.AA16635@milk10> Date: 19 Apr 88 21:15:36 GMT Sender: daemon@ucbvax.BERKELEY.EDU Organization: The Internet Lines: 38 Last year I asked some questions about TCP sockets under 4.2-4.3BSD UNIX, concerning a dynamic alternate routing capability. Specifically I was concerned that under 4.2 routes are closely tied to the TCP sockets using that route, such that if the route dies then the TCP peers at the sockets also time out and die, forcing the application using the sockets to restart the sockets. Rumor had it that 4.3 would solve some of these problems, because the TCP timers could last long enough for a new route to be obtained. The part of the problem which was perhaps not addressed under 4.3 was the caching of only a single 'good' route at a time. Now I have heard that the Sun Internetwork Router (based on XNS) under the newer versions of SunOS supports load sharing, which leads me to believe that Sun must also support redundant routes in the routing tables. I have a need for highly responsive route switching under the socket layer in the work I'm doing, where the physical layer tends to be poor or non- existant on a per-route basis over time. If the functionality has not been implemented yet I have an interest in trying it myself, but I don't want to re-invent the wheel, Sooo... My questions are: (a) Has Sun implemented what I call here "dynamic redundant routing"? If so, under which releases of the OS? If so, is it only for the Internetwork Router? (b) If Sun has not implemented it, has Berkeley, or someone else? (c) If Sun has not implemented it, how are they doing load sharing over the I.R.? Did I hear a wrong rumor? Please call, write or post to the net. Thanks, Robert Allen, robert@spam.istc.sri.com 415-859-2143