martin@sabre.UUCP (martin levy) (02/20/85)
I am looking for some clues as to the usage of ARP requests to deal with routing in IP networks. i have heard about it and would like to hear if anyone can give me some more leads. What ones does is not bother to check if the IP address is on local network and just send out an ARP request anyway. Then the correct gateway will send an ARP responce back. With this scheme there is no need to have routing code in 'workstations', you only need to have in in gateways. any pointers (or code) would be useful. Martin Levy Bellcore. PS. should there be an RFC on this, plase can someone send it.
mogul@Shasta.ARPA (02/22/85)
> I am looking for some clues as to the usage of ARP requests > to deal with routing in IP networks. i have heard about it and > would like to hear if anyone can give me some more leads. > > What ones does is not bother to check if the IP address is on > local network and just send out an ARP request anyway. Then the > correct gateway will send an ARP responce back. With this scheme > there is no need to have routing code in 'workstations', you only > need to have in in gateways. > RFC917 ("Internet Subnets", October 1984, by yours truly) briefly describes the use of ARP as a stopgap measure to avoid having to implement subnet code in all hosts. This is not the most efficient way of doing things, but it's been used successfully here at Stanford. RFC925 ("Multi-LAN Address Resolution", Jon Postel) describes a more complex scheme, where ARP replaces all subnet routing mechanisms, even between subnet gateways. Such a scheme is in use at CMU, but has several drawbacks. Neither of these RFCs addresses the use of ARP for long-distance routing, but there is no obvious reason why it wouldn't work. A new set of RFCs in in the works that will attempt to clarify all of this; I expect the people writing them will be done within the next few weeks. Among other things, there will probably be a new ICMP "Gateway Request/Reply" pair that will allow a host to find a gateway using a single broadcast message. Also, the proposal is to have gateways send only "Host Redirects", not "Net Redirects", thus insulating hosts from details of Class A/B/C addresses and subnets. Using ARP for routing, as I said, is a stopgap. In very simple code, such as a bootstrap loader, it is the appropriate choice. However, you lose the ability to work around a dead gateway, so in even minimally sophisticated "workstation" code you would do better to use a simple, per-host routing table, and the new "Gateway Request" ICMP. There's also some resistance to the "layer-crossing" required to implement ARP routing support in gateways, although I think people will bite the bullet and do it. Still, ARP is not going to solve problems for hosts on anything but a 10Mbit Ethernet.