An Efficient Hybrid Peer-to-Peer System for Distributed data Sharing

1. An Efficient Hybrid Peer-to-Peer System for Distributed data Sharing Min Yang and Yuanyuan Yang BhargaviGedala

2. Peer to Peer(P2P) Overlay network • P2P network is a logical overlay network on top of a physical network. • Each peer corresponds to a node in the P2P network and resides in a node in the physical network. • The links between peers are logical links each of which corresponds to a physical path in the physical network.

3. Structured P2P overlay network • Peers are connected by a regular topology. • Build a distributed hash table(DHT) on top of the overlay network. • DHT supports efficient data insertion and lookup. Advantage: • Efficient and accurate services Disadvantage: • Effort needed in maintaining DHT. Its vulnerable to frequent peer joining and leaving also known as churn.

4. Unstructured P2P overlay network • Organize peers in a arbitrary network topology. • Use flooding or random walks to look up data items. Advantage • Resilient to frequent peer joining and leaving. Disadvantage • Not an efficient network. Data query efficiency and accuracy is a problem.

5. Objective of the Paper

6. Related Work • Content Addressable Network(CAN). • Chord . • Gnutella. • BitTorrent. • Yappers. • A hybrid peer to peer network

7. System Overview Core transit networks and many stub networks are the 2 parts of hybrid peer to peer network. T-network organizes peers into a ring. Peers in s-network are called s-peers except for the t-peer attached to this s-network. System parameter ps.

8. Hybrid peer to peer system basic idea. • Data item is represented by a pair: key and value. • A peer uses 2 operation: store and lookup. • In hybrid system t-peers divide a range of d_id into segments and s-networks are responsible for each segment.

9. T-Peer Join/Leave • Each peer points to a successor and a predecessor. • A join request is sent to the t-peer with its p_id after obtaining a arbitrary t-peer. • Request is forwarded until the joining p_id is between a t-peer p_id and its successor. • Generation of the p_id by the server. • When a t-peer leaves the system , in a hybrid peer-to-peer system can reduce some updating new finger pointers. • Handling abrupt leaving t-peers.

10. S-Peer Join/Leave • S-peer maintains list of its neighbor and adds a random peer by acquiring its IP address to its list. • Data lookup is through flooding in Gnutella style peer-peer network. The range of flooding is determined by search radius , that is TTL value of the packet. • Constraints added in choosing a random peer when a new peer joins are: Random peers to be picked to only t-peers. Restriction on the degree of peers.

11. When an s-peer leaves the system. • Abruptly leaving s-peers are handled by sending periodic “Hello” messages. • Lot of traffic is generated by sending “HELLO” messages between peers. Acknowledgement messages are sent in response to data query message can over come the above problem. • Maintaining topology is important in structured peer to peer networks so peer crash can be detected either by acknowledgement messages or “HELLO”

12. Concurrent Join/Leave • The performance of the system can be degraded by concurrent joining and leaving of the peers. • T- peers concurrency handling is complicated as it involves 3 pairs and constraint is strict on topology. • Concurrency handling in the database system for concurrent joins or leaves in the t-network.

13. Data Insertion/Lookup • Data insertion and generation is done by the peers. • The peer hashes the key into space first when generating the data item. • If the d_id lies within the range of the current s-network and if the d_id does not lie in the range of the s-network • For look up it hashes the data key to obtain the d_id. • In data look up there is a timer to receive the data item for d_id lying within the range of the current s-network and d_id not in its current s-network range.

14. Performance Analysis 1.Performance of the Join Operation. The time difference between a joining peer request to join to the server and the joining peer being inserted is join latency and the efficiency of join operation is measured on it. II.Performance of the data lookup Operation. The time difference between the time a data lookup request is issued by a peer and the time it received by it. When the data is not received before the timer expired then the lookup operation is a data lookup failure.

15. (a) The average join latency under different s. (b) The average lookup latency under different s.

16. Enhancements Supporting Link Heterogeneity • Usage of link capacities can be maximized by connecting many lower link capacities with higher link capacity in system. • T-peer are assigned higher link capacities and the s-peers lower link capacities.

17. Supporting Topology Awareness • Logic links are made up of one or more physical links. • Topology of the overlay network may be different from the physical network. • Topology mismatch causes the performance to degrade . • Clustering the peers with respect to the latency. Interest Based s-Network • The s-peers can indicate their interest while sending a join request to the server. • Its useful also useful for partial/keyword search as well as exact search.

18. Bypass Link • s-peer can be interested in several types of data at the same time. • By pass links reduces the burden of the t-network as well. BitTorrent Style s-Network • t-peer acts as a tracker.

19. Simulation Results • The performance of the system through NS2 simulators. • Data Distribution: The t-peer can store the data in its database or insert in its random neighbor’s database (a)-(f) Probability density functions of the number of data items per peer under different ps for two data placement schemes.

20. Lookup Failure Ratio: ratio of Number of failed data lookups to Total number of data lookups The lookup failure ratio is compared by tuning the TTL and ps. and assumed that the data are looked up after insertion. • Lookup failure ratio under different TTL values. (b) Lookup failure ratio when peers crash under different ps values.

21. Lookup Efficiency: It is evaluated in terms of time and space. Connum – The bandwidth is approximated by the number of peers a lookup may contact. (a) and (b) Average lookup latency under different ps values.

22. Conclusions and Future Work • The hybrid peer-to-peer network can utilize the efficiency of the structured peer-to peer network and the flexibility of the unstructured peer-to-peer network and maintain a balance between them. • How to design a caching scheme for the hybrid peer-to-peer system.

23. Questions