IPng Paths not taken worth revisiting Jon Crowcroft

IPng: Paths not taken worth revisiting Jon Crowcroft, http: //www. cl. cam. ac. uk/~jac 22

IPng is > 20 years old, why now n IPv 6 is sort of out there n n n But not really addressing core challenges But the IPng requirements were broader More recent Future Internet programs n n n havn’t gained a lot of traction Perhaps because smart money was elsewhere Cloud, data center, apps, smart phones etc

IPng alternatives nack then…. n SIP v. PIP n n Nimrod (& BGP) n n Simple & FTIFs Maps v. Path Vector Catnip n Pluralisms (viz Plutarch)

Circumstances not what they were n Scale – yes, n n Mobile/wireless – no n n No real ip-level seamless handover Cloud, virtualisation n n roughly as predicted Not at all predicted Computation, Storage n Somewhat under-estimated

IPv 6 v. Loose Source Route/soft circuit n One of the age old debates n n Maybe with slices/vpn, can re-visit n n n Brittleness v. predictability But then what for? Route cache/fwd table lookup opt? Maybe when 100% of planet covered n n n Three times over Can start to see long term stability? Including diurnal/timezone variance etc

Maps v. Link State or Path Vector n Nimrod (+to some extent PNNI routing) n n n Maybe people were afraid of scale n n n Based on exchange of maps Rather than vectors or LSAs Or revelation of internal details To competitors or attackers Especially if combined with source route But can mitigate with accountable IP And maybe other tricks…. see later

Multi-protocol diversity n Virtualised switches n n n Many functions Fast instantiation Per flow or customer etc Partitioned… Programmable NICs n n Offload v. bypass (extreme case rdma fpga brainwave )

Low Latency and Resilience n Need cooperation between Aps n n At spectrum, wireless, coding, packet level But we’re competing in these levels too “Coopetition” Looks a bit like problem BGP solves n n n But fine grain, faster And BGP doesn’t converge or operate fast enough or describe right resources Exchange (encrypted) resource maps?

Exchange Encrypted Maps n And cooperating nodes have keys n n Paths could be braided for resilience n n Non-coop nodes use MPC to compute paths MPC run on coop nodes or use homomorphic crypto perhaps multipath Braiding can be generalised n n n Resource pool spectrum, access lines, cell towers backhaul, etc

Many apps don’t need paths n Generalise virtual path to virtual tree n n Virtual braided tree from net>customer n n Hose model+braiding For media/cast Virtual braided tree from customer>net n For Io. T, collector nets

Virtualised switches&NICs n Can run resource pool algorithms n n n Over heterogeneous links, virtual paths i. e. recursively over layers too (c. f. John Day) Braiding roughly like bundles of pathlets Virtual path/braided treelets

Re-combination of Ipng ideas n n n From Nimrod & Catnip updated to use virtualisation & crypto To get isolation and cooperation

How can we build this from now? n First steps design performant crypto n n Needed anyhow for commercial reasons Get buy-in from others requires openness By definition, this doesn’t work unless inclusive Advantage for Huawei? n n Leadership! First mover etc