User Route Choice: promoting real-time ISP
competition
or
Is Traffic Engineering Evil?
| Jon Crowcroft | |
| WAGN |
| "Some of the dynamic price work assumes that users can choose routes. | |
| In fact, in today's internet, routes are pretty much fixed by ISPs. | |
| At each level there exists a highly structured market of access, transit and core ISPs with rules and so on. | |
| Users can select access routes only by dialing different ISPs, or on a slower timescale (timescale is the critical question) can pay to have DSL, cable modem or other links installed by other ISPs. |
| But the choices are still very limited, and are not available on the TCP session or RTT like times. | |
| Recent observations of BGP indicate that a number of large client sites (typically server farms) are trying to use "more specific prefixes" as a way tp punch routes through alternate ISPs. | |
| This is wreaking a certain amount of havoc in the ISP community, mainly due to shortcomings in the BGP routing architecture. | |
| What this talk is about is the design space for user selection in routing. | ||
| There are a number of ways of carrying out loose or strict source routing, | ||
| classical IPv4 source route options, | ||
| dynamic NATs, | ||
| IPv6 hop-by-hop options, | ||
| application layer relays and so on. | ||
| Each has its shortcomings, and the time is about right for offering input to medium routing research in the IETF/IRTF, to get a solution to efficient selection of "articulation" points to drive inter-ISP competition. | ||
| Efficient Route Selection | |
| Without Route computation | |
| Expose metrics and conditions | |
| Not over-complicate packet processing |
| Can ISPs or users cheat? | |
| Nope….not for long… | |
| Same reasons you can’t cheat for ECN – incentives aren’t there for ISP or user to cheat – | |
| Even easier to detect route hackery than ECN hackery strategies…as more globally visible |
| Is tricky! | |
| See YAM or QoSMIC for receiver leaf selsection protocols | |
| I guess for PIM SM, RP allocation could be part of scheme – for inter-domain, need to worry about MBGP/MSDP (would be nice if we had BGMP and or H-bidir-PIM!) |
| GSE? | |
| Like multi-homing, but with handover- only problem might be route update frequency (see forwarding overhead) and re-computing compressed route prefix/label replace tables… | |
| Maybe a problem for convergence… |
| Easy…same as striping I nnormal case –treat as a bundle of connections at different prices | |
| Or maybe just bill average route price? | |
| Could be nice solution to some problems of justifying “protection” Bandwidth that is idle (I.e. we sell it cheapJ |
Impact on IP forwarding/route?
| Hum – GSE v. NAT? | |
| Router already compute ttl– and re-run checksum | |
| Incremental longest match, prefix/label swap and checksum should be similar | |
| Problems: traceback? |
| Need to consider enforcement – M3I model is to do this via money and admission (per packet, per flow, per aggregate) – maybe other permits could be allowed | |
| Need to implement hierarchical monitoring | |
| Risks need understanding |
| BGP is being busted | |
| IP is being busted | |
| Just like IP busted telcos… | |
| So lets fight back and fix IP and routing… | |
| Just like we fixed TCPJ |
| N==3 ? | ||
| Can pick ingress, egress AND core – this is important as it levels the playing field for core (tier 1/0) provicers | ||
| Egress is important for client and server side resilience. | ||
| Use BGP Community Attribute – need #AS^2 worth of 3 values – diameter of net is 7, so 50*3 – fits in 1 byte…. | ||
| M3I has a metering/edge admission architecture | |
| Need route choice mechanism | |
| Addr space hack works with GSE with IPv6 – there doneJ | |
| Need a break… |