roode@orc.olivetti.COM (David Roode) (04/16/88)
The April 1988 issue of Data Communications magazine on p. 78 has an entry called "Kludgey NSF RT-based network?" It says it turns out each NSFNET T1 backbone switch will consist of from 3 to 9 RT's linked via a token ring. Local campuses will interface to one of the RT's via Ethernet and backbone trunks will be handled by multiple RT's. A dedicated RT will handle routing. The article then goes on to quote an unnamed 'key university network manager' who bemoans the 115 amps of current the switch will consume and the 32,500 BTU/h of cooling required. The article says power consumption was confirmed by the NSFnet operators, but says they referred the issue of heat dissipation to IBM.
Mills@UDEL.EDU (04/17/88)
The reporter called me a month or so ago with the same quote. I told him that, even if the quote were correct, it had nothing to do with the functionality or performance of the net and was probably quoted out of context. Consider this: it could have been worse in the case of some other publications known to me. In the past I have had tailfeathers singed myself with publications quoting me out of context, so I have some sympathy for the quotee. Geeze, is it really 115 Amps? Dave
wesommer@athena.mit.edu (William Sommerfeld) (04/18/88)
In article <8804170013.aa05021@Huey.UDEL.EDU> Mills@UDEL.EDU writes: >Geeze, is it really 115 Amps? Well, if you believe what's printed on the back of the boxes, the IBM Reduced Taxes[1] desktop model is rated at 7.5 amps at 120VAC. The floor model is rated at 10 amps. The floor model has more expansion slots as well as room for up to three disk drives; the desktop model only has room for one internal disk. With 9 CPU's, that would be 67.5 or 90 amps total. RT's tend to run hot; you're not supposed to run them for very long (more than 5 minutes or so) with the cover off, or else they reportedly melt down. The CPU chips are mounted in massive heat sinks which are directly in the blast of a 4" fan. Four RT/PC's tend to keep a 10' x 20' room nice and toasty warm... - Bill [1] so called because it is rumoured that IBM gives more of them away than they sell...
berger@clio.las.uiuc.edu (04/19/88)
Would anybody care to compare the old ARPA IMP boxes to this scheme?
Mike Berger
Department of Statistics
Science, Technology, and Society
University of Illinois
berger@clio.las.uiuc.edu
{ihnp4 | convex | pur-ee}!uiucuxc!clio!bergerkwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (04/21/88)
I found the Data Comm article intriguing. I haven't heard anything on usenet or the internet regarding the new NSFnet router architecture. Could someone point me toward a journal article, trade rag article, or other source where I could learn something about how these new RTs route? If there is nothing in print, I would like to invite someone from MERIT or other knowledgable developer to post something about how these RTs work compared to some plain vanilla (:-) router like a cisco, a proteon, a fuzzball, or homegrown router. Any takers? Kent England, Boston University
Dave_Katz@UM.CC.UMICH.EDU (04/21/88)
In response to Kent England:
?> If there is nothing in print, I would like to invite someone
?> from MERIT or other knowledgable developer to post something about how
?> these RTs work compared to some plain vanilla (:-) router like a
> cisco, a proteon, a fuzzball, or homegrown router.
The routing algorithm internal to the backbone is an implementation of
the ANSI IS-IS Intradomain routing proposal. In a nutshell, it is an
SPF variant with stable link metrics (assigned by "system management,"
which could be human or electronic). Reachability of routers and
networks are flooded throughout the backbone.
EGP is used to pass reachability info between the backbone, the
regionals, and the ARPAnet. Various filtering tricks are used to
try to ensure that ARPA routes don't leak out into the regionals,
for example. Tables of potential EGP announcements are used to
ensure that bogus info is not imported from the regionals.
The routing algorithm is documented in ANSI X3S3.3/87-150R, which
we will be making available for anonymous FTP once the NSFnet
Info Services machine is in production. Local adaptations of
the algorithm to DOD IP and the EGP mechanisms will likely be
published in a paper once the panic subsides a little.
Dave Katz, Merit/NSFnetfedor@NISC.NYSER.NET (Mark S. Fedor) (04/22/88)
Kent, Your are indeed brave! The NSFNet routing scheme (RT routing) is not for the week of heart! I could mail you relevant papers if you need them. Mark`
jsloan@wright.EDU (John Sloan) (04/22/88)
Apart from the Datacomm article, all the wide area networks, including NSFNET, got a lot of bad, or at least interesting, press in the _The_Chronicle_of_Higher_Education_, 6 April 1988, A11. The gist of the article is that the WANS (NSFNET, ARPANET and BITNET are specifically mentioned) are victims of their own success, and that they don't have the financial resources (and all that implies: staff, equipment, etc.) to support the exploding growth in their use. It mentions too the cost of local (campus-wide) area networking, and that many universities are finding it expensive to support, also in part because of the growth in usage. The _Chronicle_ is a trade rag for the education industry. University administrators read it routinely. The thing that concerns me is that they'll only see the bottom line, not the fact that most of these problems are occurring because the networks are immensely popular. I just co-wrote a proposal for a more extensive campus network, and this article came out right when we fielded the proposal to the administration. Great. In times when university administrators are more bean-counters than educators (and, admittedly, perhaps necessarily so) this is probably going to make selling networking more difficult. -- John Sloan, The SPOTS Group Wright State University Research Building CSNET: jsloan@SPOTS.Wright.Edu 3171 Research Blvd., Kettering, OH 45420 UUCP: ...!cbosgd!wright!jsloan +1-513-259-1384 +1-513-873-2491 Logical Disclaimer: belong(opinions,jsloan). belong(opinions,_):-!,fail.