[homenet] Off topic [was: Layering [was: Despair]]
On 09/08/2015 09:04, Geoff Thompson wrote: ... The success of the (reasonably well layered) TCP/IP suite would indicate that the market has decided that this is a cost well worth paying. Precisely my point, except that it is not true. The datagram service that was provided for with such success by TCP/IP does not provide the same service over all physical layers. In fact the now predominant physical layers do not provide sufficiently low-jitter, low loss service for all legacy services to work well. You mean, sufficient compared with 4800 baud modems over spotty analogue phone lines, which were predominant when those legacy apps, right up to HTTP/1.0, were invented? What don't work well are *modern* services invented for broadband, in the absence of anything accurately described as broadband. But, yes, I hope the final list of requirements for the homenet routing protocol includes: works adequately on lossy wireless media with poor multicast support. Brian ___ homenet mailing list homenet@ietf.org https://www.ietf.org/mailman/listinfo/homenet
[homenet] Multicast over 802.11 and Babel [was: Multicast in IPv6]
I just find it strange that you have hit the multicast problem for routing protocols but not for IPv6. Ah, I now understand that we were speaking about different things. The issue that Babel used to have with 802.11 multicast was not packet loss -- Babel is designed to be extremely resilient to packet loss, it handles 80% loss before breakfast. The issue was about collisions. 802.11 uses CSMA/CA, which is not as good at avoiding collisions as the familiar CSMA/CD (which cannot be used on wireless). Since multicast packets are sent at a glacially slow rate, multiple nodes multicasting simultaneously can cause the network to collapse. In earlier days, engineers from a company that wishes to remain anonymous were testing Babel by putting 1500 mesh routers, tuned to a single frequency, in a room, and booting them all simultaneously. They found that when doing that Babel would cause massive amounts of collisions for no less than 20 minutes; the network worked fine once the bootstrap had finished. The issue was fixed by doing two things: rate-limiting replies to wildcard requests and adding massive amounts of jitter to urgent TLVs (those that the spec says MUST be sent in a timely manner). Of course, this increases convergence time on wired networks, but something had to give. There are two lessons to be drawn from that experience: 1. don't put 1500 wifi routers in a single room; 2. making a routing daemon that works well in a variety of conditions is hard work, and requires large amounts of careful testing. But please don't take my word for it -- it would be way more helpful if you could do your own testing in order to find out how well IS-IS works over wireless, and publish your experimental results. As far as I am aware, nobody has done that, so this would be a seminal paper in its (admittedly narrow) niche. -- Juliusz ___ homenet mailing list homenet@ietf.org https://www.ietf.org/mailman/listinfo/homenet
Re: [homenet] Off topic [was: Layering [was: Despair]]
Brian- On Aug 9, 2015, at 1:19 AMPDT, Brian E Carpenter brian.e.carpen...@gmail.com wrote: On 09/08/2015 09:04, Geoff Thompson wrote: ... The success of the (reasonably well layered) TCP/IP suite would indicate that the market has decided that this is a cost well worth paying. Precisely my point, except that it is not true. The datagram service that was provided for with such success by TCP/IP does not provide the same service over all physical layers. In fact the now predominant physical layers do not provide sufficiently low-jitter, low loss service for all legacy services to work well. You mean, sufficient compared with 4800 baud modems over spotty analogue phone lines, which were predominant when those legacy apps, right up to HTTP/1.0, were invented? Actually, I do. First of all, it wasn't that spotty in terms of errors per hour (which is what you need to count to factor out data inflation) and it had a number of secondary characteristics that weren't taken into account in formulating the layered model. What don't work well are *modern* services invented for broadband, in the absence of anything accurately described as broadband. Not really. Communicating time information works much better in a real-time circuit switched network than it does in a packet based store and forward network. But, yes, I hope the final list of requirements for the homenet routing protocol includes: works adequately on lossy wireless media with poor multicast support. Brian Geoff ___ homenet mailing list homenet@ietf.org https://www.ietf.org/mailman/listinfo/homenet
