hedrick@TOPAZ.RUTGERS.EDU (Charles Hedrick) (11/03/86)
Thanks. I had in mind doing it as a device driver that pretended to be point to point links. It's hard to see why this would have to know anything about DECnet. As long as the bits get there, I find it hard to see how DECnet could know the difference. The only question I see is whether the interface between my device driver and the DEUNA device driver can be made to work, particularly in the presence of the Wollongong code. My suspicion is that it might work only when no other IP was hapneing for that interface. (On a number of machines, we are now using separate interfaces for IP and DECnet.) The problem, of course, is figuring out when an IP or ARP packet gets handed to Wollongong and when to my hack. I intentionally suggested using point to point links because there is no structure to them. If I tried to emulate DECnet support of Ethernet, I would have to handle multicasts and make the system think it saw an area router. I don't know about the X.25 support, but assume there are protocols involved to open connection that would have to be interpreted. I am to meet with someone from DEC this week on this issue. Apparently at least someone is interested in looking into it. I find that DECnet is currently my biggest networking headache. I can get good TCP implementations for every machine other than the VAX. For the VAX we have things that are either incomplete or the Wollongong thing whch is too expensive for large-scale use and still doesn't handle mail right. I know a number of people who think DEC wants it that way. That's always hard to judge. But the only convenient way to build a campus network that will pass DECnet is to use level 2 routers. We are very reluctant to do this because of concerns about Ethernet meltdown being propagated around the campus. Stevens is building a campus network with LANbridges. But the person responsible didn't know about the problem with Ethernet meltdown. They referred me to the DEC person who is doing their design, and he didn't seem to care. I suggested maybe a simple filter to prevent broadcasts from passing when the packet type is IP. He would rather die than suggest that DEC should do anything based on packet type, since that is non-ISO. The problem with depending upon migration to ISO is that it looks like the last piece of ISO to fall into place will be the network layer, and that is precisely what I need. In fact, the only plausible strategy for implementing ISO that I have heard is to layer it on top of IP.