N ETWORKed M EDIA L AB. D EPT. OF I NFO. & C OMM., K-JIST DHT Based P2P (Peer-to-Peer) for Exploiting Network Proximity ChanMo Park [email protected]

NETWORKed MEDIA LAB.

DEPT. OF INFO. & COMM., K-JIST

DHT Based P2P (Peer-to-Peer) for Exploiting DHT Based P2P (Peer-to-Peer) for Exploiting Network ProximityNetwork Proximity

ChanMo [email protected]

Jan. 9, 2004


DEPT. OF INFO. & COMM., K-JIST 2/20

ContentsContents

Problems

Estimating Internet Distance

Exploiting Topology Information

Approaches towards exploiting network proximity

Discussion



ProblemsProblems

P2P File Sharing DHT (Distributed Hash Table) Based Approaches

o CAN ( ’01 - ACIRI)

o Chord (’01 - MIT )

o Tapestry (’01 - Berkeley)

o Pastry (’01 - Microsoft) What can help P2P File Sharing?

o Performance of DHT based P2P Routing (Lookup and Insert) Nearest node as Next hop Exploiting Network Proximity



Estimating Internet DistanceEstimating Internet Distance

Triangulated Heuristic ’00 USC IDMaps ’01 University of Michigan GNP (Global Networking Positioning) ’02 CMU Vivaldi ’03 MIT (Chord)



Triangulated Heuristic ’00 USCTriangulated Heuristic ’00 USC

Select N nodes in the network to be base nodes

A node H is assigned coordinates (N-tuples) which are the distances between H and the N base nodes

Then the distance between H1 and H2 is bounded below by

bounded above by U, L, or (U+L)/2 can be dista

nce estimation

Base nodes

A

B|)(|max21,...,2,1 ii BHBHNi ddL

|)(|min21,...,2,1 ii BHBHNi ddU



IDMaps ’01 Univ. of MichiganIDMaps ’01 Univ. of Michigan

Special HOPS servers maintain a virtual topology map of the Internet consisting of end hosts and Tracers

Distance of host A and B is estimated as D=a+b+c

AB

Tracer 1

Tracer 2

ab

c



GNP (Global Network Positioning)’02 - CMUGNP (Global Network Positioning)’02 - CMU



Vialdi’03 - MITVialdi’03 - MIT Distributed algorithm assigning synthetic coordinates in D-

dimensional space to Internet hosts Predicting latency between two hosts by Euclidean

distance between their coordinates Coordinate of each node

o Simulating node’s position in a network of physical network Sampling the network latency between each node and a few

other nodes, adjusting the node’s coordinates to minimize the error between the predicted and sampled latencies.

No fixed infrastructureo Collecting latency information from only a few other hostso Piggy-backing latency information on application traffic



Exploiting Topology InformationExploiting Topology Information

Proximity Routing Proximity Neighbor Selection Geographic layout



Proximity RoutingProximity Routing

Proximity routing is when the routing choice is based not just which neighboring node makes the most progress towards the key, but is also based on which neighboring node is closest in the sense of latency.

compromise between progress in the ID space and proximity in routing

Adv.

o Light weight, easy to implement

o Workable for all algorithms Limitation

o Improvement depend on number of neighbor.



Proximity Neighbor SelectionProximity Neighbor Selection

Proximity criterion is applied when choosing neighbors, not just when choosing the next hop

Topology constructing Routing Table Adv.

o Improve performance for prefix routing Limitation

o Can Not apply to CAN, Chord, etc.



Geographic layoutGeographic layout

Geographical layout is to choose node identifiers in a geographically informed manner.

Choose nodeID according to physical network Adv.

o Successfully apply to CAN Limitation

o Not apply to one dimensional ID space overlay (Chord, Tapestry, Pastry, etc)

o Need landmark

o Destroy uniform ID populationLoad balance



Approaches towards exploiting network proximityApproaches towards exploiting network proximity

1. Efficient Topology-Aware Overlay Network ’02

2. Efficient Topology-Aware Overlay Network ’02

3. Exploiting network proximity in peer-to-peer overlay networks ’02

4. Topology-aware routing in structured peer-to-peer overlay networks ’02

5. Topologically-Aware Overlay Construction and Server Selection ’02

6. Building Topology-Aware Overlays using Global Soft-State ’03



Efficient Topology-Aware Overlay Network ’02(#1)Efficient Topology-Aware Overlay Network ’02(#1)

Mithos Proximity Neighbor Selection Efficient routing

o Cartesian coordinate systemo Irregular hypercube mesh

Locality-preservingo Chose neighbors wisely

Routing and forwardingo Quadrant-basedo Only neighbors knowno Efficient forwarding

Assuming 2-D

o Connect to each quadrant

o Forward according to destination vector

o Routing becomes trivial



Efficient Topology-Aware Overlay Network ’02 (#2)Efficient Topology-Aware Overlay Network ’02 (#2)

Node ID cannot be known a priori

Find out during join phase Descend towards global

minimum Stabilize using "spring forces" Result: Node ID



Exploiting network proximity in peer-to-peer overlay networks ’02Exploiting network proximity in peer-to-peer overlay networks ’02 Topology-aware routing in structured peer-to-peer overlay networks Topology-aware routing in structured peer-to-peer overlay networks

’02 ’02 Pastry (Tapestry) Proximity Neighbor Selection Using mechanism to build routing tables taking network proximity

into account Proximity Invariant

o Each entry in a node X’s routing table refers to a node that is near X, according to the proximity metric, among all the live Pastry nodes with the appropriate nodeID prefix



Topologically-Aware Overlay Construction and Server Selection ’02Topologically-Aware Overlay Construction and Server Selection ’02

Based on CAN Distributed Binning

o Require a set of landmark node spread across the Interneto Select a particular bin by Measuring its distance to a set of landmarkso Using Landmark orderingo Partition the coordinate space into m! equal sized portions



Building Topology-Aware Overlays using Global Soft-State ’03Building Topology-Aware Overlays using Global Soft-State ’03

eCAN Landmark clustering + RTT

o Computers with similar landmark vectors are likely to be close

o measures RTT to top candidates to find the nearest neighbor

o Use landmark vector as the DHT key

U

Landmark vectorU=(u1,u2,u3,u4)

L1 L2 L3 L4Landmark

nodes

u1 u2 u3 u4

V Landmark vectorV=(v1,v2,v3,v4)

v1 v2 v3 v4

U’

Landmark space

Landmark vectorB

A



DiscussionDiscussion



ReferencesReferences

1. Zhichen Xu, Chunqiang Tang, and Zheng Zhang: "Building Topology-Aware Overlays using Global Soft-State". The 23rd International Conference on Distributed Computing Systems. May 19-22, 2003 Providence, Rhode Island USA (CAN Proximity Neighbor Selection (Landmark + RTT))

2. S. Ratnasamy, M. Handley and R. Karp, “Topologically-Aware Overlay Construction and Server Selection”, Proceedings of Infocom, 2002(CAN Geographic Layout)

3. Miguel Castro, Peter Druschel, Y. Charlie Hu, and Antony Rowstron. “Topology-aware routing in structured peer-to-peer overlay networks“, In FuDiCo 2002: International Workshop on Future Directions in Distributed Computing. University of Bologna Residential Center Bertinoro (Forli), Italy, June 2002. (Pastry Proximity Neighbor) Selection

4. Exploiting network proximity in peer-to-peer overlay networks. Miguel Castro, Peter Druschel, Y. Charlie Hu, and Antony Rowstron. Technical report MSR-TR-2002-82, 2002 (Pastry Proximity Neighbor Selection)

5. Marcel Waldvogel and Roberto Rinaldi, “Efficient Topology-Aware Overlay Network”, Hot Topics in Networks I (HotNets-I), October 2002.The HotNets-I proceedings will also appear in Computer Communication Review, Volume 33, Number 1, January 2003, http://www.cs.washington.edu/hotnets/papers/waldvogel.pdf ,Delaunay triangulation

Documents

N ETWORKed M EDIA L AB. D EPT. OF I NFO. & C OMM., K-JIST DHT Based P2P (Peer-to-Peer) for Exploiting Network Proximity ChanMo Park [email protected]