We went back to SIGCOMM 1995, when one of us (Bruce) was chair of the inaugural Outrageous Opinion Session. We remember some highlights of that session, but we had forgotten that Bruce was also scheduled to give a tutorial, competing for attention with Marc Andreesen. Marc cancelled at the last minute citing "business pressures" so a bunch of disappointed people came to Bruce's session. 2/n
In the outrageous opinions session, David Clark from MIT gave a memorable talk entitled "We should all become economists". He applied this advice himself, with a good example being his paper "End to End arguments vs. the Brave New World" https://groups.csail.mit.edu/ana/Publications/PubPDFs/Rethinking%20the%20design%20of%20the%20internet2001.pdf
Unfortunately, in our view, not too many people took his advice in full, although plenty of tech people did decide to become armchair economists. 3/n
Paul Francis, co-inventor of NAT, made the point (controversial at the time) that just saying no to NAT would work about as well as telling teenagers to abstain from sex. Fortunately the position on NAT has shifted and we now deal pretty well with its prevalence (see NAT traversal). 4/n
Finally, there were a surprising number of talks on the merits of IP vs. ATM, on both sides of the argument. Hard to believe that ATM ever had such a bright future (in the eyes of many) given that it now occupies a small space in the set of technologies that sit underneath IP. It turned out that we didn't need fixed length packets to build high speed networks! Full story in our newsletter: https://open.substack.com/pub/systemsapproach/p/outrageous-opinions?r=cxpek&utm_campaign=post&utm_medium=web /fin
Lots of political pull for ATM back then. It mildly impacted our group's internal visibility at NRL in the mid-90s when the ATM folks had much louder megaphones. It was quite a pain in the ass.
(Our NRL group did one of the first IPv6 implementations and to our knowledge the first IPsec (AH/ESP only) implementation.)
@danmcd@SystemsAppr The issue with ATM was that it was unclear if you could get enough quality of service with IP to avoid jitter. The answer, as it turned out, was to build very fast networks, ones that handled enough streams of data that the average behavior was good enough. But "very fast" wasn’t easy—I have a 1994 book “Gigabit Networking” on my shelf now, about the challenges it posed. Now, of course, it's routine on LANs and WANs are very much faster. But it wasn't obvious back then.
@SteveBellovin@danmcd@SystemsAppr Craig's book? That was a great one at the time - gigabit seemed so exhilarating when we were only just getting fast ethernet (100Mbps) and you could still find coax hanging around.
@dave_andersen@danmcd@SystemsAppr Definitely coax around then. In the Firewalls book, I mentioned cutting the transmit leads on a monitoring device, to make sure that it never sent packet that an attacker could detect. That is, I cut the transmit leads on the 15-pin drop cable to a transceiver in thick coax. You can't do that with twisted pair…
@SteveBellovin@dave_andersen@danmcd@SystemsAppr On the jitter part -- very fast may not have been easy, but it was very clear that the Internet was getting faster quicker. The rule of thumb was that various QOS tricks allowed you to carry 20% more load with the same jitter, but that if capacity doubled every year it really did not matter.
it was kinda happening in places but definitely wasn't consistent everywhere. at the time that the NSF was funding 45mpbs, we (uunet) were still selling new dialup and 56k.
we did go from 56k/fractional T1 to OC3 in the 6 years i was there for at parts of the backbone.
customers and over water links were a different story.
@paul_ipv6@huitema@dave_andersen@danmcd@SystemsAppr If I recall correctly, there was another use of ATM, too: UUnet—arguably the first major ISP, now part of Verizon Business—used ATM as its layer 1/2 transport. That let them reconfigure their topology at lower layers to match and balance load.
Add comment