From mboxrd@z Thu Jan 1 00:00:00 1970 From: ggm@algebras.org (George Michaelson) Date: Tue, 12 Dec 2017 12:17:39 +1000 Subject: [TUHS] V7 Addendem [ really lawyers and AT&T consent decree ] In-Reply-To: <20171212020432.4821118C08C@mercury.lcs.mit.edu> References: <20171212020432.4821118C08C@mercury.lcs.mit.edu> Message-ID: It lives on, in the QoS tagging gamers and VOIP people do on their home routers, in a faint hope that apart from outbound queues, inbound queues from their local provider do some kind of (re)prioritization based on it. Since thats the person who they pay money to, it kind-of makes sense. I didn't mean to disrespect the people who did the models or the protocol or standards work btw. Like you, I think it was a solution in search of a problem, in a point in time now past. What we have now, is a horrid war on capital investment. Nobody wants to turn up the unlit glass, because it would expose the pricing models which depend on artificially constructed scarcity. It interests me that a lot of stuff which doesn't work 'in the wide' does work in specific contexts. So I would not be surprised if RSVP and like things survive inside large corporate networks. In like sense, aircraft databusses are often just normal switches with isochronous TDM slot markers, to give them guaranteed/bounded delivery behaviours. I think Christian Huitema did some stuff . on that in Ethernet while he was in INRIA (or somebody in the same unit) -G On Tue, Dec 12, 2017 at 12:04 PM, Noel Chiappa wrote: > > From: George Michaelson > > > I don't think this list is the right place to conduct that particular > > debate. > > Not disagreeing; my message was a very short gloss on a very complicated > situation, and I wasn't trying to push any particular position, just pointing > out that work (whether the right direction, or not, I didn't opine) had been > done. > > > Its true RSVP didn't get traction, but the economics which underpin it > > are pretty bad, for the current Internet model of settlement > > Yes, but would _any_ resource reservation system, even one that _was_ > 'perfect', have caught on? Because: > > > it would not surprise me if there is ... more dropped packets than > > strictly speaking the glass expects. > > This is related to something I didn't mention; if there is a lot more > bandwidth (in the loose sense, not the exact original meaning) than demand, > then resource reservation mechanisms buy you nothing, and are a lot of > complexity. > > While there were bandwidth shortages in the 90s, later on they pretty much > went away. So I think the perception (truth?) that there was a lot of headroom > (and thus no need for resource reservation, to do applications like voice) > played a really big role in the lack of interest (or so people argued at the > time, in saying IntServ wasn't needed). > > Noel