Supporting Complex Multi-dimensional Queries in P2P Systems Bin Liu, Wang-Chien Lee Hong Kong University of Science and Technology ICDCS ‘05 Dik Lun Lee Pennsylvania State University
One line comment The system supports multi-dimensional range and kNN queries in P2P network considering peer heterogeneity
Motivation Complex Queries for emerging P2P applications E.g. P2P auction network for real estate - Find available buildings at most 10km from the airport (range) - Select three available buildings closest to the airport (kNN)
Problem Existing P2P systems cannot support such complex queries Can, Chord, Pastry, Tapestry, etc – not for range queries Mercury, Squid, etc – only for range queries
Solution Approach CAN w/o hashing NR-tree indexing Range Query, kNN Query DHT: routing P2P system Queries
Solution Approach S1 P1 P2
Indexing in a Cluster abcdefghijkE4E5E6E7E8 E1E2 y-axis x-axis P2 E1 E2 E5 E4 E6 E7 E8 a b c d e g h i j k f NR-tree (residing in S1) S1
Join (Leave/Failure) abcdefghijkE4E5E6E7E8 E1E2E3 y-axis x-axis P2 E1 E2 E5 E4 E6 E7 E8 a b c d e g h i j k f E3 NR-tree (residing in S1) S1 P3 S1 P2 P3 P4 S2
Query Processing – Range query abcdefghijkE4E5E6E7E8 E1E2E3 y-axis x-axis P2 E1 E2 E5 E4 E6 E7 E8 a b c d e g h i j k f E3 NR-tree (residing in S1) S1 P3 query q S1 S3 S2 S4 S5
Query Processing – kNN query abcdefghijkE4E5E6E7E8 E1E2E3 y-axis x-axis P2 E1 E2 E5 E4 E6 E7 E8 a b c d e g h i j k f E3 NR-tree (residing in S1) S1 P3 query q S1 S3 S2 S4 S5 E2E1E3
Analysis of the system Routing cost Dynamic load balancing is difficult Join 할 때 한 peer 가 관리하는 data 에 locality 가 있다는 가 정이 현실성 부족 Not efficient kNN query support Query result correctness 를 100% 보장 못함 Node heterogeneity 를 고려했지만 어떠한 보장도 못함
Experiment Parameters clusters 12 nodes
Experiment – cost vs. distribution per query per node join/leave
Experiment – cost vs. query size
Experiment – cost vs. cluster size
Experiment – TTL
Experiment - # of backup super-peers