Endless Arguments in Systems Design Jon Crowcroft, University of Cambridge Currently IMDEA Networks. Jon.Crowcroft@cl.cam.ac.uk http://www.cl.cam.ac.uk/~jac22 GENI – Global Environment for Network Innovation US NSF/Extra Money Calls for Intl participation Planetlab++ (Intel/300 site) FIND – Future Internet Design NSF long term research in NeTS programme 40+ Projects Quite mixed bag [EU Trilogy project (which UC3M and UCL and others are in) did analysis of them] email WWW phone... The Internet SMTP HTTP RTP... Protocol Hourglass TCP UDP… (Deering) IP ethernet PPP… CSMA async sonet... copper fiber radio... Putting email WWW phone... on SMTP HTTP RTP... Weight TCP UDP… IP + mcast + QoS +... ethernet PPP… • requires more CSMA async sonet... functionality copper fiber radio... from underlying networks Mid-Life email WWW phone... Crisis SMTP HTTP RTP... TCP UDP… • doubles number IP4 IP6 of service interfaces ethernet PPP… • requires changes CSMA async sonet... above & below copper fiber radio... • major interoper- ability issues Or is GENI losing its bottle? Network Innovation is a Hot Topic Internet problems abound 2/3G problems abound Convergence driving incremental Critical Infrastrure driving radical Many US GENI/FIND +EU +Asia projects There appear to be some ok ideas But also some wrong headedness This is a case study on wrong headedness Incremental from Convergence SCTP (PSTN signaling transport on IP) SIP&NAT traversal&VOIP in general IPTV (Multicast & P2P) BGP fast Convergence Traffic Engineering (whether OSPF TE or RSVP or MPLS based) Neutrality fights Long Term IP Service Best Effort So DDoS not meaningul Not helpful – need architectural change Fundamentally, Mobile, Multicast, Multihome Possibly Multiprotocol (esp. Control Plane) DoS Resistence, built in Possibly different economic/charging models Need shift in model Maybe not a stack anymore Certainly not IP ++ ++ ++ i.e. not just IPv6 with current IPv4 knobs and warts Two wrongs that don’t make a right(#) Virtualisation of Networks Next Generation Wireless Data Driven Networks (#) Two parts of FIND/GENI work in US that I feel they are doing wrong and we are doing right + one thing I think the US and EU are both doing right. Veni, Vidi, Vici, but not Vini? Virtualisation is not about VPNs(*) Hierarchical Hard Multiplexing won’t scale VPNs are an ok idea in the Enterprise Its too slow to reconfigure on outages Its bad for resource allocation in wireless C.f. hard spectrum allocation C.f. hard channel allocation * Not just VPI/VCI (for those who recall B-ISDN!) Virtualisation of Control not Data In Xen/Xorp work: key is about dynamics and isolation of functionality, not fwding performance Need to deploy disjoint service ctl True, vpn style IP fwding is useful too But no unified way of doing this on wireless Or on multipath Or for netcoded data Or swarms… Next Gen Wireless Cooperative Diversity in all things Antennae, coding, routing, power, spectrum, “channels” Behavioural constraints enforced (aside: F# smart phone:-) Infrastructure still there for services, e.g. To provide radio map (with memory) Security/Identity/Location Assistence Maybe even payment Control plane for virtual wireless communities Data Driven, Declarative, We’ve seen 1-1, 1-n, n-1 comms patterns Now we’re seeing 0-n and n-0 US loves to re-invent(+) - but in fact Pub/sub;event/notify isn’t all there is: Think swarm Think anysource multicast Exaflood is no big deal when decentralised…properly Think exploding google (aside: haggle:) (+) architecture is not just creative children’s stories. Virtual, Wireless, Declarative Data? We’d all like to get rid of wires (on the other hand, we’d probably all like to get rid of batteries too:) So high capacity applications should be user contributed content too What is the control plane for these to co-exist with each other? C.f. nano-datacenters and haggle, unified 0wning half a PC (or hub or STB) Significant fraction of internet access via Home Hubs Significant fraction of wireless acces via smart phones Home hub as PVR, smart phone as PC Lots of storage + faces 3 ways Faces “wide area”; faces home Faces neighbours… … … Changes economic relationships completely Provider coexist w/ virtual providers If provider wants to scale net, Run P2P for CDN on home hub And do end run on data centers Let home user also run P2P CDN Let wireless user also run community and ad hoc mesh Virtualise STB + Handsets With suitable declarative(pub) Ctl Plane Paradigm shift now enabled Revenue model from services not bitpipes Way forward? Devil is in detail Remove separation of end + intermediate system All systems may have storage Must restore flow, congestion and reliability control functions to “network” But do not need them “point-to-point” So they are going to look very different Everyone is a DTN custodian & there is no “network” Summary/Conclusions GENI is ok for next generation core IP Need more fundamental changes if we address high speed wireless everywhere Especially since application paradigm already shifting Europe well placed to contribute: both wireless and application architecture, which in turn contribute to immediate and long term network architecture better than incremental core work…. not that that is a bad thing either!
Pages to are hidden for
"Research is communication"Please download to view full document