MAP@LCS.MIT.EDU (03/20/89)
Date: 19 Mar 89 08:14:40 GMT From: vsi1!daver!dlr@ames.arc.nasa.gov (Dave Rand) [Dave asks:] Why do machines ARP for broadcast IP addresses? Because they have two bugs! First, they think they should be forwarding packets (as if they were a gateway, even though they aren't). Second, they have a mistaken impression of the broadcast address and so don't recognize it and think it's just another host. This is because one of the more popular bases for networking code came out during a time when the spec on broadcast addresses was in flux. The way the problem manifests itself is that when one host broadcasts a packet using the right address, any hosts with these two bugs will believe it's for a specific host and attempt to forward there. They will ALL send out an ARP request for this address AT THE SAME TIME! This gives lots of collisions on these hosts. Lots of older systems work this way. The best way to keep this traffic down is to find these systems and upgrade them (provided the vendor supports up-to-date code). The second best (and often more practical) approach is to use a non-spec broadcast address that all your hosts will at least recognize. The spec address for broadcast is ones, what you'll probably find is that the deficient hosts only take zeros but that the "correct" hosts can be configured to use this as well. __ /| /| /| \ Michael A. Patton, Network Manager / | / | /_|__/ Laboratory for Computer Science / |/ |/ |atton Massachusetts Institute of Technology Disclaimer: The opinions expressed above are a figment of the phosphor on your screen and do not represent the views of MIT, LCS, or MAP. :-)