Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!wuarchive!decwrl!mogul From: mogul@decwrl.dec.com (Jeffrey Mogul) Newsgroups: comp.protocols.tcp-ip Subject: Re: Directed broadcasts?? Message-ID: <312@jove.dec.com> Date: 28 Feb 90 00:51:27 GMT References: <11545@nlm-mcs.arpa> Organization: DEC Western Research Lines: 33 In article <11545@nlm-mcs.arpa> gish@host.UUCP () writes: Would someone please outline how to perform a directed broadcast-- that is, a broadcast to a net located one or more hops away? Are there severe limitations or restrictions on the use of directed broadcasts, such that directed broadcasts are impractical? e.g., are routers usually configured to block off-site broadcast packets? Well, you might look at RFC922: Mogul, J.C. Broadcasting Internet datagrams in the presence of subnets. 1984 October; 12 p. (24832 bytes) On the other hand, I think it is nearly universally agreed that if broadcasting is evil, then directed broadcasting is evil compounded. I.e., I wouldn't suggest using them (and as far as I know, few routers implement directed broadcasting). BTW: the reason for asking this question (aside from my naivety) is that I am investigating providing network services on machines whose network identities may change (although the subnet probably wouldn't); and I'd like to avoid having to update potentially thousands of distributed copies of client applications when these changes do occur. I'd be very interested if perhaps y'all have some better suggestions than directed broadcasts. Sounds like you want to use multicasting (see RFC1112 and RFC1075). But you might have to wait a while for that to be widely implemented. Or maybe you should just use the Domain Name System (DNS) to provide some late binding (i.e., assign a nickname to the host that is currently providing the service, and require the clients to look up that nickname whenever they want to use the service). You can do it today if your software uses DNS (and if it doesn't, fix that first). -Jeff