morgan@jessica.stanford.edu (RL "Bob" Morgan) (11/24/88)
Since no one else has commented on it, I thought I'd voice some approval of the improvement in service that I've seen since the NSFNet has come on-line in a big way in the last few months. When inter-campus traffic (between here at Stanford and CMU, for example) went via ARPANet, I would never see better than about 2.5 KBps (that's thousands of, er, octets per second 8^) throughput for FTP, even at night. During the day it was generally worse or unusable. Now, even in midday I regularly get rates of 12-15 KBps, and the initial connection sequence is much snappier. Given that the slowest link between here and many places is now a 64 KBps T1-divided-by-3 virtual circuit, I guess there's still room for improvement, but it's still a Big Step Forward. Kudos to all concerned. Onward to Gigabits. - RL "Bob" Morgan Networking Systems Stanford
steve@NOTE.NSF.GOV (Stephen Wolff) (11/24/88)
Washington bureaucrats like me take unwarranted pleasure in unsolicited compliments like yours, when the credit really goes (in this particular instance) to all the hard-working and dedicated folk at Merit, Inc. (who have a five-year agreement with NSF for management and operation of the NSFNET Backbone), and to their joint study partners, IBM and MCI, whose contributions of facilities, services, high-grade engineering talent, and top management attention and commitment have consistently exceeded anybody's reasonable expectations. I'd also like to note here the continuing and welcome participation of the Strategic Fund of the State of Michigan in this rather special academic-industrial-state- federal partnership. But, warranted or not, I'll just <grin>. Tnx! -s (Networking, NSF)
kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (11/24/88)
In article <4210@Portia.Stanford.EDU> > morgan@jessica.stanford.edu (RL "Bob" Morgan) writes: > > >Since no one else has commented on it, I thought I'd voice some >approval of the improvement in service that I've seen since the NSFNet >has come on-line in a big way in the last few months. Not only is throughput up. I see much greater stability in routes and *no more black holes*!! The network tables from jvncnet are *huge* and still growing quite rapidly, but things stay stable. The link state (SPF) algorithm is proving itself. I still think the hardware is an oddity in terms of actual cost, power consumption, etc, but the software and technical expertise is first rate. Hats off to Hans-Werner, the MERIT crew, and the IBM and MCI guys. You want the same stability for your own nets? Then demand an open SPF algorithm from your router vendor.
hwb@MERIT.EDU (Hans-Werner Braun) (12/04/88)
Kent: Thanks to you and to Rob Morgan for the kind comments. Let me say, though, that while for the stability of the network the IBM implementation of the ANSI IS-IS protocol plays a major role, the real issue is the overall routing design. In particular what, as far as I can tell, made a, if not the, difference was that we do not translate metrics between multiple administrative domains any more. For example, from a backbone point of view, only up/down events for network announcements from the regionals matter. The backbone does not care about whether links within a regional switched so that the regional, e.g., RIP metric jumped from five to six. Also the connections, routing wise, are more engineered into primary and secondary considerations. While the whole routing announcements are dynamic (IGP in the regionals, ANSI IS-IS in the backbone and EGP between the two) the allowed routing announcements and their mapping into primary and secondary considerations are controlled by configuration files. This is currently still a people-time intensive undertaking, but we are in the midst of automating a good part of the configuration setup. To be fair it should also be said that in the previous incarnation of the NSFNET backbone (the 56Kbps Fuzzball based system) the resources to do all this work were simply not available. Some IDEAS (I think IDEA20 and IDEA21) as well as RFC1074 describe the NSFNET routing to some extend. Many people contributed to the current NSFNET working in the way it does. We are planning and working on further improvements and should have more information about this soon. We are trying to get some summary into the next Link-Letter. Rob said something like "on to gigabit networks." That won't be cheap as getting bits across the country at that speed is not a small expense. Even moving from a T1 network to 45Mbps (T3), as the Merit NSFNET proposal suggested to do during the duration of this project, will require additional resources. -- Hans-Werner Date: 24 Nov 88 15:11:31 GMT From: kwe@bu-cs.bu.edu (kwe@bu-it.bu.edu (Kent W. England)) Organization: Boston Univ. Information Tech. Dept. Subject: Re: NSFNet throughput applauded Sender: tcp-ip-relay@sri-nic.arpa To: tcp-ip@sri-nic.arpa In article <4210@Portia.Stanford.EDU> > morgan@jessica.stanford.edu (RL "Bob" Morgan) writes: > > >Since no one else has commented on it, I thought I'd voice some >approval of the improvement in service that I've seen since the NSFNet >has come on-line in a big way in the last few months. Not only is throughput up. I see much greater stability in routes and *no more black holes*!! The network tables from jvncnet are *huge* and still growing quite rapidly, but things stay stable. The link state (SPF) algorithm is proving itself. I still think the hardware is an oddity in terms of actual cost, power consumption, etc, but the software and technical expertise is first rate. Hats off to Hans-Werner, the MERIT crew, and the IBM and MCI guys. You want the same stability for your own nets? Then demand an open SPF algorithm from your router vendor.