ZBEN@umd2.ARPA (Ben Cranston) (07/18/86)
> X-From: "Roger Fajman" <RAF@NIHCU> > Date: Wed, 16 Jul 86 19:54:06 EDT > There are also useful differences in routing outside of a single > site. For example, my node (NIHCU) is connected to the UMDD at the > University of Maryland, only 10-15 miles away from here. They also > have an ARPANET connection and the BITNET and ARPANET machines are > themselves connected. So, if I want to send something to one of the > ARPANET machines at the University of Maryland, it doesn't make a > whole lot of sense for me to send it a thousand miles via multiple > network hops to WISCVM, just so it can turn around and come all the > way back through ARPANET. I don't have an ARPANET connection, so I > can't just put it onto ARPANET. Yet, sending all my ARPANET traffic > through the University of Maryland ARPANET connection isn't > necessarily right either, as they may well not want to or be able to > handle all of it. Roger and I have talked about setting up something to do this. My main qualm at this point is that my connection to the Internet (ARPAnet) is currently through the Department of Computer Science's mimsy, which is a very busy machine, and I fear the possible political consequences of suddenly foisting off a much greater packet traffic on them. We have our Internet connection basically at their sufferance. When last we talked I suggested to Roger that we get together and discuss this matter in the future, when we had our own Internet connection. At the time this was slated for June 5, but it is now July 18 and it has not happened yet. Sorry Roger, you know how schedules slip. Not my fault. My other concern is network bandwidth. Many of our connections are actually as slow as 9600 baud, and significant traffic could cause bottlenecks that might impact our operation. The umd2-umdd link was upgraded from 9600 to 64kb, but the umd2-mimsy link is still 40kb, and the umdb-gwuvm link is still 9600. I don't know about Roger's link, he is probably on umdb but I don't know the bandwidth offhand. ----------------- But his point is a good one. If there is only one defined gateway it is going to be a bottleneck. There needs to be some way of spreading out the load. Here at UMD we have an exit for xxx.UMD.EDU pointing at UMD2 so mail for mimsy.umd.edu, cvl.umd.edu, etc goes through the on-campus gateway. But if somebody says mimsy.arpa or cvl.arpa then it's WISCVM time again. The same problem can occur under nameserving. Let's say Roger registers blue.NIH.GOV and I request nameservice. If I read the Internet RFC stuff correctly all it can do is return WISCVM, or whatever the defined gateway to the RSCS transport service might be. It might return me multiple MX resource records, but they will be prioritized, and the standard says I must try them in priority order. I have no choice about maybe using an on-campus gateway in preference to WISCVM. I am not sure from the standard if the nameserver is allowed to look at *my* Internet address and maybe tell me something different than it would tell *you*... Anybody feel confident enough to comment on this?