7
Comparing Hybrid Comparing Hybrid Peer-Peer Systems Peer-Peer Systems Beverly Yang Beverly Yang Hector Garcia-Molina Hector Garcia-Molina Stanford University Stanford University Presented by Kalyan Boggavarapu Presented by Kalyan Boggavarapu

Comparing Hybrid Peer-Peer Systems Beverly Yang Hector Garcia-Molina Stanford University Presented by Kalyan Boggavarapu

Embed Size (px)

Citation preview

Page 1: Comparing Hybrid Peer-Peer Systems Beverly Yang Hector Garcia-Molina Stanford University Presented by Kalyan Boggavarapu

Comparing HybridComparing Hybrid Peer-Peer Systems Peer-Peer Systems

Beverly Yang Beverly Yang Hector Garcia-MolinaHector Garcia-MolinaStanford UniversityStanford University

Presented by Kalyan BoggavarapuPresented by Kalyan Boggavarapu

Page 2: Comparing Hybrid Peer-Peer Systems Beverly Yang Hector Garcia-Molina Stanford University Presented by Kalyan Boggavarapu

Different Modals [1]Different Modals [1]

Page 3: Comparing Hybrid Peer-Peer Systems Beverly Yang Hector Garcia-Molina Stanford University Presented by Kalyan Boggavarapu

Hybrid modal[1]Hybrid modal[1]

Page 4: Comparing Hybrid Peer-Peer Systems Beverly Yang Hector Garcia-Molina Stanford University Presented by Kalyan Boggavarapu

► Eg: NapsterEg: Napster► Currently we study: Currently we study:

OpenNapOpenNap► New issues: IR in P2P, New issues: IR in P2P,

co-operative web co-operative web cachingcaching

► Modeling peer-peer Modeling peer-peer systems [1]systems [1] Query modeling, Query modeling, performance modeling, performance modeling, validation with real data validation with real data

OpenNapOpenNap

Query model [1]

f – query “selection power”

g – query popularity

Correlate f and g with exponential distribution

Page 5: Comparing Hybrid Peer-Peer Systems Beverly Yang Hector Garcia-Molina Stanford University Presented by Kalyan Boggavarapu

ArchitecturArchitecturee► LoginsLogins: batch & incremental: batch & incremental

► Types of architecture (partly Types of architecture (partly [1]) [1]) :: Chained – forward a queryChained – forward a query

► Large and diverse data not many Large and diverse data not many servers.servers.

► Dis: expensive if many servers have Dis: expensive if many servers have to satisfy the queryto satisfy the query

Full replication – make a copy of Full replication – make a copy of indexes on every server.indexes on every server.► Dis: login information must be sent Dis: login information must be sent

to every serverto every server Hash – holds subset of index of Hash – holds subset of index of

wordswords► Bandwidth consumption would be Bandwidth consumption would be

more.more. Unchained- NapsterUnchained- Napster

► High data redundancy, Many serversHigh data redundancy, Many servers► Dis: restricted access to files, they Dis: restricted access to files, they

cannot see the user files in other cannot see the user files in other systems.systems.

Query model assumption [1]: Most popular query has the largest selection power.Assumptions for the experiment [1]:

•Servers are connected via LAN

•Users and servers are connected via same LAN network

Experimentation:

Hybrid architecture, OpenNap

Page 6: Comparing Hybrid Peer-Peer Systems Beverly Yang Hector Garcia-Molina Stanford University Presented by Kalyan Boggavarapu

Comparing P2P systemsComparing P2P systems

Conclusions: [1]• Incremental login scales better than batch login • Unchained architecture results in lower number of results returned• Hash architecture is resource intensive• Full replication works well when result sets are large• Chained architecture is well-suited for music-sharing, but suffers from poor query performance

References:[1] http://db.uwaterloo.ca/~tozsu/courses/cs856/presentations/jinxiao.pdf

Page 7: Comparing Hybrid Peer-Peer Systems Beverly Yang Hector Garcia-Molina Stanford University Presented by Kalyan Boggavarapu