CS 268: Project SuggestionsNext Two LecturesOverviewOutlineRED: Does it Really Help?Quickstart+TCP vs XCPBurst SwitchingInterdomain Traffic EngineeringRedirection/Selection StrategiesSlightly Nontraditional NetworkingDisjoint Paths vs WaypointsResiliency via IncastNegation RoutingSensornet RoutingReconfigurable Directional AntennaeAnycast as Evolution MechanismDistributed Hash TablesDHTs OverviewExample: ChordLocation Control in DHTsContent Routing (DHTs?)Circle Checks for DHTsDHT as Library vs DHT as ServiceOpenDHT Projects!!!Peering DHTsNew Architectures and ParadigmsDoS PreventionCheckable ProtocolsAnnotation LayerTheoryCAP vs CASOverlay RoutingGeographic RoutingIdentity-Based Architectures“ID-based” Architectures“ID-based” Architectures (cont’d)Example realization in i3 and DOAAuthentication and EncryptionSignaling Protocol for MiddleboxesID based Transport ProtocolsService Differentiation via MiddleboxesAnonymous File SharingEvent Notification SystemAnycast / Service LocationEfficient Multi-Level NamingInternet-Scale XML Dissemination Service [Due to Yanlei Diao – next 5 slides]ONYX: Large-Scale XML DisseminationContent-based RoutingContent-based Routing with I3?Dissemination of ResultsNext StepCS 268: Project SuggestionsScott Shenker and Ion StoicaJanuary 24, 20052Next Two LecturesWednesday:-Clark: “The Design Philosophy….”-Saltzer, Reed, and Clark: “End-to-end arguments…”Monday:-Cerf and Kahn: “A Protocol for…”Remember to do your summaries!Reading list will be finalized over weekend….3OverviewWill present ~35 short project suggestionsLegend: based on how well-defined projects are, not necessary how difficult they are - Well-defined projects- Less-defined project- You need to define project’s goalsNeed to send us a one page proposal by Feb 7-Feel free to talk with us beforehand!4OutlineTraditional networkingSlightly nontraditional networkingDistributed Hash TablesNew Architectures and ParadigmsTheoryIdentity-based Architectures5RED: Does it Really Help?Random Early Drop is the first and most widely used “active queue management” algorithm.Its goal is to promote fairness and decrease queue lengths (delays).Does it really help? There have been several contradictory papers on this.What is the real story?6Quickstart+TCP vs XCPXCP (Katabi et al.) is a recent congestion control proposal (we’ll cover it later) that requires dramatic changes in TCP and routersQuickstart is a quick-and-dirty hack:http://www.icir.org/floyd/quickstart.htmlIs XCP significantly better?7Burst SwitchingTwo main communication models-Datagrams: each packet is individually switched (routed)-Circuits: a circuit is set-up and all packets are forwardedHybrid model: burst switching-First packet describes how many packets are in a burst-Router decides whether to forward all packets in the burst or none of themResearch-Design a burst switching protocol and study its trade-offs8Interdomain Traffic EngineeringInterdomain traffic engineering is a mess:-Ambiguous goals-Ad hoc techniquesThe best known paper on this is "Guidelines or Interdomain Traffic Engineering" by Feamster et al.Can one come up with a specification language and a coherent set of mechanisms?9Redirection/Selection StrategiesAkamai redirection uses domain information to redirect requestsRecent work on server selection uses network coordinates (GNP, Vivaldi)Are coordinates significantly better?Slightly Nontraditional Networking11Disjoint Paths vs WaypointsFeedback based routing (Zhu et al.) uses disjoint paths to achieve resiliencyWould using waypoints work better?-Easier (no need to find disjoint path)-More choices12Resiliency via IncastSend to set of waypoints (in mcast group):Each waypoint forwards data toward rcvrIncast boxes (one or more along path) strip out extra redundancies (configurable parameter)How reliable does that make delivery?What is a coherent architecture for this?-i3, DOA, etc.?13Negation RoutingRecent proposal for Packet Obituaries (Argyraki et al., Hotnets 2004) gives feedback on which AS is dropping packetsAdditional proposal (being written up) allows one to do “negation routing” by saying: “avoid this AS”What is the performance of this approach?14Sensornet RoutingPoint-to-point routing is hard in sensornets and other ad hoc networksIn a static ad hoc network, one can build up a coordinate system by using recursive pairingChallenge: design such an algorithm and analyze its performance15Reconfigurable Directional AntennaeLots of interest in “mesh networking”-Many performance problems because of interferenceWhat if we had reconfigurable directional antennae instead of broadcast?Could quickly reconfigure “links” to produce good pathsDesign such a system and analyze it16Anycast as Evolution Mechanism[joint with Sylvia Ratnasamy]How can the Internet evolve?Need to give incentives for individual ISPs to deploy new versions of IP at least partiallyThat requires having packet using IPvX being automatically forwarded to the nearest router supporting IPvX Interesting combination of technical and economic requirementsDistributed Hash Tables18DHTs OverviewEach data item and machine (node) in the system has associated a unique ID in a large ID spaceHash table like interface-put(id, data)-data = get(id)ID space is partitioned among nodesData items are stored at the node responsible for its ID19Example: ChordCircular ID space [0..2m-1]Consider two consecutive nodes on the ID circle with IDs N1 and N2, where N2 follows N1-Node N2 is responsible for interval (N1, N2]Node 35 inserts (37, data) Node 3 queries data with ID 373720354137 dataChord circle02m-120Location Control in DHTsUsers have no control over where data items are locatedAdvantages: -Users don’t need to know about individual nodes-System can recover in case of failure without user involvementDisadvantages:-Not efficient-Enforces uniform trust modelResearch: design a system in which users have “some” degree of control on where data items are located -E.g., “I want my items to be located only on nodes in US”21Content Routing (DHTs?)Gritter and Cheriton proposed a technique for “Content Routing”This was before the days of DHTsWould DHT technology (put “on-path”) provide a better solution to this
View Full Document
Unlocking...