evaluation guess and non-forwarding peer-to-peer search icdcs paper beverly yang patrick vinograd...

30
Evaluation GUESS and Non- Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford Un iversity

Upload: thomasine-sims

Post on 06-Jan-2018

216 views

Category:

Documents


0 download

DESCRIPTION

By 溫健順 3 Introduction The key to the usability of P2P  To search for and retrieve efficient DHT-based search technique  Focused on the availability  control the placement of data and the topology of the networks

TRANSCRIPT

Page 1: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

Evaluation GUESS and Non-Forwarding

Peer-to-Peer search

ICDCS paper Beverly YangPatrick VinogradHector Garcia-MolinaComputer Science Department, Stanford University

Page 2: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 2

Outline Introduction GUESS Protocol GUESS vs. Gnutella Policies Maintaining the Link Cache Conclusion Gain and Question

Page 3: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 3

Introduction The key to the usability of P2P

To search for and retrieve efficient DHT-based search technique

Focused on the availability control the placement of data and

the topology of the networks

Page 4: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 4

Introduction part 2Forwarding-based

Forwarding until the stop criterion is met

Difficulty To control how many peer receive

the message Which peers receive message

Page 5: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 5

Introduction part 3Non-forwarding search

Peers directly probe each other

Under GUESS protocol

Page 6: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 6

Introduction part 4The focus point of paper

When processing the query In what order should peers be

probed

Page 7: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 7

GUESS Protocol part 1Basic architecture

Two cache, or lists of pointer (IP address) Link cache Query cache { IP address of Q, TS, NumFiles,

NumRes }

Page 8: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 8

GUESS Protocol part 2Maintaining state

Make sure entries in the cache are fresh

Ping message Pong message Introduction protocol

Page 9: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 9

GUESS Protocol part 3Query propagation

Iteratively probe, unicast query

After sending , peer either receive the reply or wait for a timeout period

Page 10: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 10

GUESS vs. Gnutella part 1 Query performance

Gnutella The number reached is determined

by the flooding GUESS

Peer has control the order and decides how many peers to probe

Page 11: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 11

GUESS vs. Gnutella part 2 State maintenance

Gnutella Consistent of active network connection.

GUESS Each peer maintains the pointer caches Cache size limited by memory. No explicit notification when new peer

joins in, it need the probing to know the new peer.

Page 12: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 12

GUESS vs. Gnutella part 3 Security

Selfish Gnutella

Fairly robust GUESS

Could easily probe more peers Imposing a higher load than the

necessary

Page 13: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 13

GUESS vs. Gnutella part 4Security

Malicious peers Gnutella

Vulnerable to denial of service (DoS)

GUESS Cause the heavy network traffic

Page 14: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 14

GUESS vs. Gnutella part 5Fragmentation attack

Gnutella Highly-connected peers are

attacked GUESS

Malicious peers propagate their identities to the network, and suddenly disappear.

Page 15: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 15

Policies part 1 Determining how the entries or

der are used and maintained. Five type of policies

QueryProbe QueryPong PingProbe PingPong CacheReplacement

Page 16: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 16

Policies part 2The implemented policies

Random (Ran)Most Recently Used (MRU)Leased Recently Used (LRU)Most Files Shared (NFS)Most Results (MR)

Page 17: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 17

Result part 1 Maintaining the Link Cache

Page 18: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 18

Result part 2 Maintaining the Link Cache

Page 19: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 19

Result part 3 Maintaining the Link Cache

Page 20: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 20

Result part 4Flexible Extent

Page 21: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 21

Result part 5Query Efficiency

Page 22: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 22

Result part 6Query Efficiency

Page 23: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 23

Result part 7Query Efficiency

Page 24: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 24

Result part 8Unsatisfied Queries

Page 25: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 25

Result part 9 fairness

Page 26: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 26

Result part 10Capacity Limits

Page 27: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 27

Result part 11Capacity Limits

Page 28: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 28

Conclusion To achieve very efficient query

performance MR policy policy presents the

best tradeoff between efficiency and robustness.

First-generation implementation of GUESS would be to use the MR policy.

Page 29: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 29

Future workHow to make the protocol

adapt to changing network condition.

How to define against selfish and malicious peers

Page 30: Evaluation GUESS and Non-Forwarding Peer-to-Peer search ICDCS paper Beverly Yang Patrick Vinograd Hector Garcia-Molina Computer Science Department, Stanford

2004.12.1 By 溫健順 30

Gain and QuestionNetwork sizeConcept of cache

ContentSize