i claim the world has changed since the reasons multicast failed - within which i include software failures (twice, cisco's implementation of multicast broke unicast across large sections of the internet - but this is ancient history) - more recent in network compute (whether it is P4 in data centers, or just SDN controller stuff in general, means that three of the major obstacles (in the Sprint paper by diot et alk: address management, ddos prevention, billing) are surmountable. actually, if one was to pursue Scion, or some other accountable internet protocol, then two of these would be trivial, even for Any Source Multicast (i.e. multicast classic:-)
but i'm still optimisitc about IPv6 too:-) On Wed, Nov 30, 2022 at 9:49 AM Dirk Trossen <dirk.tros...@huawei.com> wrote: > > Hi John, > > Thanks for the comments. Please see inline. > > Best, > > Dirk > > -----Original Message----- > From: Int-area <int-area-boun...@ietf.org> On Behalf Of John Gilmore > Sent: 30 November 2022 03:16 > To: Luigi Iannone <g...@gigix.net> > Cc: int-area <int-area@ietf.org>; g...@toad.com > Subject: Re: [Int-area] whither multicast? > > > https://arxiv.org/pdf/2211.09029.pdf > > The paper is a useful high level review of what multicast promised and why IP > multicast has failed so far. As it says, "The takeaway here is that the woes > of previous multicast deployments can be overcome by observing the lessons > learned from those deployments..." > [DOT] Glad to see that you saw useful messages in the paper. > > But its cheerful-charlie outlook about its "proposed new multicast semantic" > (which exact semantic, isn't clear) > [DOT] Being called 'cheerful' as a cynical German is quite unusual to me 😉 > The attempt was to outline the view that multicast isn't dead or not useful > anymore, on the contrary (see other drivers, even outside of where IP > multicast would be applied, in Section V). That may count as cheerful already > for some though, I admit. > > doesn't seem to address any of three major issues in how today's Internet > relates to IP > Multicast: > > * Should the allocation of hundreds of millions of IPv4 addresses to > native multicast be reconsidered? Only a tiny fraction of them are > in use, while unicast is the clear success story of the Internet. > Unicast address demand continues to rise, while multicast demand is > invisibly small. IANA has never allocated roughly half of all > multicast addresses to anyone for any use (225-231/8 and > 235-238/8). The vast bulk, of the tiny fraction that were > allocated, are for services now considered obsolete (all but LAN > use, e.g. for Neighbor Discovery or MDNS; and all but > Source-Specific Multicast, which is limited to 232/8). > [DOT] To me, this question is a consequence of what is being discussed. If a > multicast semantic was to separate WHAT from HOW, questioning the allocation > of routing identifiers is indeed valid. > > * For end-users seeking access to nonlocal IP multicast, over the last > few decades, one issue at ISPs continued to get in the way. Large > backbone unicast routers could not reliably be used to forward > native multicast traffic, because enabling multicast on those > routers tended to make them crash (more often). When seeking > 99.9999% uptime for the unicast traffic that pays all the bills, > and seeking low network debugging and support costs, having a > backbone router crash for even ten minutes a year from running a > niche service, caused ISPs too much trouble. So when multicast was > supported at all, it tended to be supported on side-machines that > were not in the main pipelines of the ISP, were not as highly > capable as the mainline routers, and required custom tunnel > configuration from each end-user -- a non-starter for mass market > use. > > * And with few or no ISPs supporting multicast by default, no end-user > applications that tried to use it ever caught on. So there was > very little demand encouraging other ISPs to enable multicast. > [DOT] Both of these last two issues are part of the chicken-and-egg problem > that often exists in technology. If demand was sufficient, striving for > matching the supply would usually happen (and being fortified in realization, > so without crashing of parts of the supply chain), while the lacking and > inadequate supply of course hampers demand. > > [DOT] This raises the question on what may break this vicious cycle. We > largely approached this from the 'more and new demand' angle. One may think > of multicast efficiencies (and the possible impact on the environmental > impact of unicast proliferation), which is outlined towards the end of > Section V with some example work; I would welcome more thoughts and > contributions for this angle, which may also relate to the soon-to-happen IAB > workshop on environmental impact. > > > The paper does not seem to address any of those issues. > [DOT] Not directly indeed - see above - but it's positioned as a working > paper in the hope to incorporate wider thinking into a revision, so your > comments are certainly useful achieving that hopefully. > > John Gilmore > > > _______________________________________________ > Int-area mailing list > Int-area@ietf.org > https://www.ietf.org/mailman/listinfo/int-area _______________________________________________ Int-area mailing list Int-area@ietf.org https://www.ietf.org/mailman/listinfo/int-area
