lmorales@natchez.huachuca-emh8.army.mil (Luis F. Morales,Jr.) (05/04/91)
I have a question regarding the routing architecture of non-MILNET networks in the Internet. What is the routing architecture in the rest of the Internet?? On the MILNET we are critically dependant on the DCA Mailbridge core gateways. As some of you may have experienced in traversing the MILNET they have been having a lot of problems over the last year. I can't imagine that the MILNET is more complex then the rest of the INTERNET but I haven't heard of any major problems on NFS-NET. Can someone please explain how routing is handled on NFS-NET for example?? Is everything statically routed? or are dynamic routing protocols used etc?? Thanks in advance, *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-* | | * Luis F. Morales,Jr. * | U.S. Army Computer Engineering Center | * lmorales@huachuca-emh8.army.mil * | | * "Expert in the field of unexplained phenomena * | and their applicability to computer systems." .......... | *-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*-*
tgsmith@fedps.East.Sun.COM (Timothy G. Smith - Technical Consultant Sun Baltimore) (05/06/91)
In article <34736@boulder.Colorado.EDU> lmorales@natchez.huachuca-emh8.army.mil (Luis F. Morales,Jr.) writes: >I have a question regarding the routing architecture of non-MILNET networks >in the Internet. What is the routing architecture in the rest of the Internet?? >On the MILNET we are critically dependant on the DCA Mailbridge core gateways. >As some of you may have experienced in traversing the MILNET they have been >having a lot of problems over the last year. > >I can't imagine that the MILNET is more complex then the rest of the INTERNET >but I haven't heard of any major problems on NFS-NET. > >Can someone please explain how routing is handled on NFS-NET for example?? > >Is everything statically routed? or are dynamic routing protocols used etc?? Since I seem to run into a question similar to yours every other month or so I will take some time and try to explain how I understand things. (NB: How I understand things does not necessarily relate to reality although I like to think I have a pretty good grasp on the history). It sounds to me like you are frustrated with MILnet's performance and reliability and are starting to ask logical questions like "If NSFnet works so well why doesn't MILnet?". I appreciate your frustration. I used to be a MILnet user when I worked for the Navy and later for the Army. I have suffered through the outages and the hassles of MILnet and know first hand how unpleasant it is. I think what you are failing to see is that MILnet is suffering from the electronic equivalent of congestive heart failure. There are just too many bits being shoved into the wires. Capacity has not been increased while usage has skyrocketed. The ARPAnet/MILnet was conceived of a long time ago primarily as network designed for connecting minicomputers to each other. This was back in the days when the computing model was a large timesharing machine with lots of terminals hung off it. Hosts were attached to ports on PSNs (aka IMPs) and the PSNs were interconnected to form the network. The lines speeds used were generally 9.6, 19.2, and 56 kbps. The only guarantee in the world of computing is that things will change and may very well change very quickly. As smaller, faster, cheaper computers came on the market LANs started popping up all over the place. Folks decided that it would be nice if they could interconnect their LANs so the got out their source code and beat on it and taught their minicomputers how to be packet switches. Suddenly the end nodes on the network weren't end nodes anymore and thus ARPAnet evolved from a network connecting hosts to a network connecting networks. Now instead of there being one computer connected to the network at each site there might be 10 or 100 or 1000. With the increase in the number of machines there was an increase in the amount of traffic that was being piped through the network and network congestion started appearing. A few years back congestion problems with MILnet and ARPAnet led to what became known as the collapse of the Internet. The collapse of the Internet was just that - data simply was not getting through the network. Two major things happened as a result of the collapse: 1) Van Jacobson studied TCP and made some major improvements to the protocol that significantly reduced the congestion. 2) People realized that the model of how the network worked had changed and that the network needed to be re-designed to reflect the new model. If everyone on the MILnet installed the modified TCP then the congestion on the network in theory have decreased and allowed the network to continue functioning albeit with lower throughput but lower throughput is better than no throughput. Number two above eventually resulted in the construction of the NSFnet and the destruction of the ARPAnet. NSFnet is designed as a system of regional networks that are interconnected. The regional networks interconnect the regional members' networks and provide connectivity to the backbone which interconnects the regional nets. The NSFnet has increased the line speed of the backbone links from the orginal 56 kbps to 1.544 Mbps (T1). I believe that there are also 45 Mbps (T3) links in place. The NSFnet has also installed new routers and new routing techonology. MILnet has basically remained static. MILnet is still primarily using 56 kbps lines, EGP, and old routing technology. I believe that DCA has actually been decommissioning circuits and gear over the last few years. The connectivity between MILnet and the rest of the world is through the infamous mailbridges. The mailbridges are generally overworked and underfed. A few years back when I was still using MILnet daily I pretty much gave up on doing interactive work over MILnet and only used it for mail as it was just to frustrating to try and use MILnet for interactive work. I think that lack of capacity and obsolete technology are what are killing MILnet. I doubt that changing routing protocols would really win much even if it were possible to get DCA and BB&N to do so (I don't feel that it is possible). Based on my experience with MILnet and its keepers (DCA and BB&N) I do not expect MILnet to Internet connectivity to get better any time soon. If you are on MILnet and need better connectivity to the Internet then I would suggest you do the same thing other MILnet sites have done - find a way to connect to the NSFnet. If you are on MILnet and need better connectivy with other MILnet sites then you should try and point out to DCA that MILnet problems are affecting your organization's abilty to execute its mission. Good luck! Tim Smith (formerly formerly of the US Naval Academy and now formerly of the US Army BRL) US mail:Sun Microsystems E-mail: 6716 Alexander Bell Drive internet:tgsmith@east.sun.com Suite 200 uucp :sun!tgsmith Columbia, MD 21046 MaBell :(301)290-1234
cjw@marmot.nersc.gov (Cathy Wittbrodt) (05/07/91)
To find out how NSFNET is routed, you might send a message
to nsfnet-info@merit.edu. I am sure they would be glad to
help you out. I can tell you that they do indeed use dynamic
routing. They also use a sort of policy routing.
Cathy Wittbrodt
National Energy Research Supercomputer Center
Lawrence Livermore National Laboratory
cjw@nersc.gov
415-422-4016