Internet Indirection Infrastructure (i3) Stoica et. al. (2002)

Selected Topics of Technological Application Areas for Structured P2P

Internet Routing Area

– i3

MANET Area

– DHT over MANETs

– Ekta

– MADPastry

– Backup Chord

– Self-organizing KBR Layer:

– Scalable, adaptive, error resilient routing

– Identifier Keys:

– Flat ID space of arbitrary size

– Hash keys can be viewed as semantic-free names (=identifiers), reducing IP addresses to locators

– Storage und Lookup:

– IDs can be dynamically created and removed (including lookup semantic) ffacilitates rendezvous processes

Internet Routing Area

Potential Contributions of DHTs

Internet Indirection Infrastructure (i3) Stoica et. al. (2002)

– Objective: Define a new Internet communication model that provides seamless support of Multicast, Anycast, Mobility, Service Composition …

– Idea: Indirect rendezvous-based approach

– Publish-subscribe-notify model on the overlay

– Communication as joint action of sender & receiver(s)

– DHT resident on i3 communication servers

– used to store and retrieve communication tuples

– Implemented & deployed on PlanetLab

– Project page: http://i3.cs.berkeley.edu/

i3 - Basic Communication

– To initiate data reception for a ‘channel’ ID, a receiver R inserts a trigger (ID,R) into the DHT.

– To send data with ‘channel’ ID, a sender S issues the packet of (ID) into the DHT

– Optimizations:

– server caching

– location-aware private triggers

i3 - Mobility

– (Public) channel IDs are position- / address- independent

– A mobile receiver simply updates its trigger(s)

– Sender mobility transparent

– Issue:

Proximity-aware private triggers invalidate

i3 – Multicast

– Simple Multicast: Triggers inserted by multiple receivers f Communication identical to unicast

– Scalable Multicast:

– Receivers use a hierarchy of triggers:

(ID, ID’):(ID’,R)

– Differs from

unicast at receivers,

but transparent to senders

– Hierarchy can be optimized using “branch-and-bound”

i3 - Anycast

– Anycast group members (receivers) insert triggers with a common prefix p

– Packets share prefix p and are delivered according to longest prefix match

– Suffixes may be used to implement policies:

– random IDs for load balancing

– location IDs for proximity matching

i3 – Service Composition

Service composition by stacking IDs f application-layer source routing:

– A sender may use (id_T, id) to send packets via T

– A receiver may use (id, (id_T, R)) to stir data through T

References

• Ion Stoica, Daniel Adkins, Shelley Zhuang, Scott Shenker, Sonesh Surana, "Internet Indirection Infrastructure," Proceedings of ACM SIGCOMM, August, 2002.

• Ion Stoica, Daniel Adkins, Shelley Zhuang, Scott Shenker, Sonesh Surana, "Internet Indirection Infrastructure," IEEE/ACM Transactions on Networking, Vol 12 (2), April 2004.

• Karthik Lakshminarayanan, Ananth Rao, Ion Stoica and Scott Shenker,

"End-host Controlled Multicast Routing", Elsevier Computer Networks, Vol 50 (6), 2006 .

MANET Area:

Conceptual Aspects

– Different concepts of (limited) lifetimes

– Overlay: Application-/User-driven

– Underlay: Mobility-/Technology-driven

– In case of independence, exp. distributed MANET lifetime:

– Topological embedding

– How to adapt to changing underlay topologies?

– Violate layers?

– Resource constraints in MANETs: Data transmission costly

(1 Transmission ≈ 1.000 arithmetic operations)

Structured Mobile P2P Networks:

Ekta (Pucha et al, 2004)

– Adaptation of Pastry to MANETs

– Integrates Multi-hop Routing Protocol into the DHT layer

– Prerequisite:

– DSR protocol in MANET

– All MANET nodes participate in Ekta Overlay

– Node-Hash applied on Underlay Address

– Achieves a late binding w.r.t. network layer routes

Ekta: Integrating the Network Layer

– Modifications of Pastry’s Routing Table and Leaf Set:

– Stores vector of Source Routes (instead of addresses)

– Route selection according to PNS (as in Pastry)

– Route employment: “Freshest among the shortest”

– Route replacement: “Least Recently Discovered”

– Source routes pushed to Ekta from network layer

(here we need coincidence of MANET : Ekta nodes)

– Network layer routes continuously monitored (from overhearing or forwarded messages)

– Prefix-based route discovery only if no route available

Ekta: Lightweight Self-Organisation

– JOIN routed to node with closest node-ID

– Closest node on reception of a JOIN:

– responds with a copy of its leaf set

– Broadcast flooding of the newly arrived node (including traversed path recording)

– Leaf set members reply with recorded path

– Unlike Pastry: Ekta abandons proximity probing

– On Departure a node floods its LEAVE

Ekta Architecture

Ekta Performance: Packet Delivery Ratio

Ekta Performance

MADPastry (Zahn et. al., 2005)

– Integrates key-based P2P routing into the network layer

– Adaptation of Pastry, combined with AODV

– Prerequisite: All MANET members run MADPastry

– Proximity exploited via random landmarking + clustering

– Idea: Forward piecewise along short, up-to-date routes

– Overlay hop count not bound by O(log(n))

Random Landmarking

– Select landmark keys (no fixed nodes)

– Nodes owning keys become temporary landmark nodes

– Clusters formed from periodic beacons (incl. hop count)

– Nodes associate to closest landmark by adopting the clusters overlay id prefix

Î Nodes change their keys!

Address Resolution

– As Nodes change keys, the overlay addressing semantic breaks

– Healing Attempt: Node, who owns initial node key operates as “address server” (Home Agent)

Î Each node has to inquire on current destination address prior to data transmission

Î On key change, nodes need to update address server

Î On address server mobility:

Address keys need to be transferred

Routing

– Degenerate Pastry Routing Table

– Contains only routes into landmark clusters

– Overlay Routing:

– According to MADPastry’s routing table (analogue Pastry)

– Underlay Routing:

– Intercept packets to discover overlay proximity, AODV-routing otherwise

– When physical route for an overlay hop is unknown:

– Cluster-broadcast packet, if in destination cluster

– AODV route discovery otherwise

MADPastry Routing

MADPastry: Performance

Hybrid Chord (Zöls et. al., 2005)

– Chord-based approach for hybrid underlays

– Presupposes subgroup of quasi-permanent nodes

– Keys only deposited on “static” nodes

– Independent of the underlay routing protocol

– Defines “Context Spaces”

– Grouping of shared objects in interest groups

– “Info Profiles” carry keywords, hashed as keys

– Suitable to optimise Boolean “AND” queries within one keyword search

HC Protocol Operations + Properties

– Nodes need additional list of static successors

– Chord key-based routing extended to next static successor

– Info profiles distributed to (multiple) static nodes according to their keyword hashes

/ Analysis & evaluation erroneous: Authors claim improvements identical to all nodes being static

Additionally: GHTs

– Geographic Hash Tables

– Use out-of-band channel for geographic location

– Routes developed according to geographically efficient paths

– Example: GCLP.

Résumé on Proposals

Ekta: Achieves late binding through layer violation

Price to pay - DHT must be present at network layer Proximity improved by extended route set

MADPastry: Late binding through packet interception in regular routing – DHT obligatory part of network layer Proximity transferred into key space

Price to pay – destroys key semantic under mobility Hybrid Chord: Key issues discarded Î static nodes

Core performance unclear

Research Issues

– Efficient late binding in overlay routes – can we leave a DHT efficient under mobility on application layer?

– How to achieve effective proximity without intercepting the logic of key-based routing?

– Is there (in some DHT) a degree of freedom exploitable for effective mobility management?

References

• C. Murthy and B. Manoj: Ad Hoc Wireless Networks, Pearson Prentice Hall, 2004.

• Nitin H. Vaidya: Mobile Ad Hoc Networks, Tutorial at InfoCom 2006, http://www.crhc.uiuc.edu/wireless/talks/2006.Infocom.ppt.

• P. Gupta and P. R. Kumar, “The capacity of wireless networks,” IEEE Transactions on Information Theory, vol. 46, no. 2, pp. 388–404, 2000.

• H.-J. Jeong, D. Kim, J. Song, B. Kim, and J.-S. Park, “Back-Up Chord: Chord Ring Recovery Protocol for P2P File Sharing over MANETs,” in International

Conference on Computational Science (2), ser. LNCS, V. S. Sunderam, G. D. van Albada, P. M. A. Sloot, and J. Dongarra, Eds., vol. 3515. Berlin Heidelberg:

Springer-Verlag, 2005, pp. 477–484.

• H. Pucha, S. M. Das, and Y. C. Hu, “Ekta: An Efficient DHT Substrate for

Distributed Applications in Mobile Ad Hoc Networks,” in Proceedings of the 6th IEEE Workshop on Mobile Computing Systems and Applications (WMCSA 2004).

References (Cont.)

• T. Zahn and J. Schiller, “MADPastry: A DHT Substrate for Practicably Sized MANETs,” in Proc. of 5th Workshop on Applications and Services in Wireless Networks (ASWN 2005), Paris, France, June 2005.

• S. Zöls, R. Schollmeier, W. Kellerer, and A. Tarlano, “The Hybrid Chord Protocol:

A Peer-to-Peer Lookup Service for Context-Aware Mobile Applications,” in Networking – ICN 2005. 4th International Conference on Networking,

Proceedings, Part II, ser. Lecture Notes in Computer Science, P. Lorenz and P.

Dini, Eds., vol. 3421. Berlin Heidelberg: Springer-Verlag, April 2005, pp. 781–

792.

• M. Wählisch: Key-based Overlay Routing in Mobilen Ad-Hoc Netzen, Seminararbeit FU Berlin, 2006.