On 7/25/14, 3:31 PM, "Juliusz Chroboczek" <[email protected]> wrote:
>RJ, > >If I understand you right, you're pushing for an approach where we don't >say anything about the routing protocol, and wait for the market to >converge on RIPng, thus ensuring interoperability. Please correct me if >I've misunderstood you. > >(My personal opinion right now (subject to change) is that we want >a single Homenet protocol, but ensure that HNCP is designed so it is >protocol-agnostic. We should make it clear that, while we encourage >experimentation with other routing protocols, deploying HNCP with >a non-Homenet routing protocol is not Homenet-compliant and doesn't give >you the right to put the cool Homenet logo on your router.) > >> Consumer electronics vendors operate with thinner margins >> (e.g. eliminating one resistor makes a meaningful difference in their >> profit). Memory footprint and CPU cost matter a great deal to these >> manufacturers. > >I think you're underestimating the amount of RAM and CPU in current cheap >routers. The three cheapest routers I can find on Wallmart's website have >between 16 and 32MB of RAM and a 300MHz MIPS CPU. That's plenty for >running RIPng and Babel, and probably more than enough for a single-area >variant of OSPF or IS-IS (Acee, can you confirm?). Yes - it should be more than enough. Recall that Markus¹ OSPFv3 Bird implementation report presented Berlin. http://www.ietf.org/proceedings/87/slides/slides-87-homenet-2.pdf I¹d expect ISIS to be similar since commercial SP implementation sizes are comparable. Thanks, Acee >(The only reason I can >see why these routers are unsuitable for Homenet is lack of flash -- some >have as little as 4MB.) > >(Note that these are too slow to push anything close to 100Mb/s, even >without firewalling, which, I suspect, is the reason why recent designs >have switched to faster CPUs.) > >> They already have RIP, > >Do they? Are there any cheap routers that come with RIP enabled by >default? > >> RIP is nearly zeroconf, and adding a link-state routing protocol would >> increase both their manufacturing cost and their support cost. >> Moreover, RIP scales just fine for residential and even small-business >> networks. > >Just like you, I really like RIP, and could certainly live with RIPng >being the One Homenet Routing Protocol. However, RIP(ng) has three major >flaws: > > - the metric is a 4-bit integer, which seriously limits your ability to > do fun stuff (like forcing traffic to follow the nice 1Gb/s cabling in > preference to the unstable wifi); > - the lack of explicit hello packets slows down link detection -- and > there's no way you'll do a cool demo unless you can unplug the ethernet > jack and have the traffic rerouted within seconds (with RIP, you'll > need on the order of a minute, unless you have link-layer indications, > which don't work well for other reasons); > - RIP is unable to detect unidirectional failures. > >I don't see the lack of source-specific routing as a serious flaw, since >desigining and implementing a source-specific extension to RIPng is >probably a job for a day or two (I'm assuming you allow me to copy-paste >from Babel and keep me furnished in coffee and food). > >> So these folks are not likely to add any link-state routing protocol to >> their products. > >While I share your distaste for the "link-state everywhere" dogma, those >folks who base their firmware on OpenWRT will add whatever doesn't require >configuration and is well supported by OpenWRT. Those who don't are going >to swear at us anyway. (Think of it as a consulting opportunity.) > >-- Juliusz > >_______________________________________________ >homenet mailing list >[email protected] >https://www.ietf.org/mailman/listinfo/homenet _______________________________________________ homenet mailing list [email protected] https://www.ietf.org/mailman/listinfo/homenet
