ClientServer versus P 2 P q Clientserver Computing

Client-Server versus P 2 P q Client-server Computing v Purpose, definition, characteristics v Relationship to the GRID v Research issues q P 2 P Computing v Gnutella, Chord, Napster, Ka. Za. A, Freenet q Summary 1

Client-Server Computing q q q Well known and powerful Server provides services and resources Multiple clients can be supported by a single server 1 -Many relationship => scalability with respect to the number of clients Model has dominated the architectural design of many applications Examples: HTTP, DNS, FTP Server Client Internet Client 2

Client-Server Computing Client-Server computing does have limitations ISSUES q Fault tolerance q Central administration q “Extreme” scalability q Unused resources in “clients” 3

Peer-to-Peer Networks (P 2 P) Peer-to-Peer computing • • • Computing paradigm where all the nodes have equivalent responsibilities and roles “neither introduces nor prohibits centralisation” “sharing of resources through direct communication between consumers and providers” “a network architecture where all the available resources are located at the network edges” “the opposite of client-server” Node Internet Node 4

P 2 P Characteristics ü Each node acts both as client and server ü Nodes are autonomous ü Network is dynamic ü There is no centralised authority (in theory) ü Network is large-scale ü Nodes have to co-operate in order to retrieve a resource or a service 5

P 2 P Benefits q Efficient use of resources q Scalability q Reliability q Ease of administration 6

Examples of P 2 P Systems Ø Napster Ø Ka. Za. A Ø Gnutella 7

Primary P 2 P research question How can we efficiently and accurately discover resources and services in a P 2 P network? Solution 1: Introduce some centralisation Solution 2: Introduce some structure “centralisation” and “structure” define two dimensions for classifying P 2 P networks 8

Types of P 2 P Systems I 9

Napster I q Sharing of music files q Lists of files are uploaded to Napster server q Queries contain various keywords of required file q Server returns IP address of user machines having the file q File transfer is direct 10

Napster II ü Centralised model ü Napster server ensures correct results ü Only used for finding the location of the files Ø Scalability bottleneck Ø Single point of failure Ø Denial of Service attacks possible Ø Lawsuits 11

Gnutella I q Sharing of any type of files q Decentralised search q Queries are sent to the neighbour nodes q Neighbours ask their own neighbours and so on q Time To Live (TTL) field on queries q File transfer is direct 12

Gnutella II ü Decentralised model ü No single point of failure ü Less susceptible to denial of service Ø SCALABILITY (flooding) Ø Cannot ensure correct results 13

Ka. Za. A q Hybrid of Napster and Gnutella q Super-peers act as local search hubs v v Each super-peer is like a constrained Napster server Automatically chosen based on capacity and availability q Lists of files are uploaded to a super-peer q Super-peers periodically exchange file lists q Queries are sent to super-peers 14

Summary q Peer-to-Peer networks are dynamic environments that facilitate resource sharing on a large-scale q Main research question is how to organise and retrieve information efficiently and accurately q Current systems use two methods: centralisation and/or structure q Focus gradually moves towards the coordinated use of versatile, distributed computing resources BUT Isn’t this what GRIDs are all about ? ? ? 15