Peer-to-peer Networks, Efficient Resource Search Technology And Its Application Research

Peer to peer network (P2P) is a revolution to the traditional Client/Server networkarchitecture. Each node in P2P can share a large amount of data with others. With the scale ofthe P2P network growing, it can rapidly accumulate inflated information and provide a richinformation warehouse for users. Consequently, an efficient search scheme for users toretrieve needed data quickly and accurately is required emergently. Present researchconcentrates on improving location efficiency, load balance, scalability and etc.P2P networks can be classified into unstructured P2P networks and structured P2Pnetworks. Unstructured P2P networks are fit for the highly dynamic network environment, forthey maintain loose network topology. But this kind of P2P network has poor scalability orconsumes too much bandwidth. Structured P2P networks employ a scalable location schemeto guarantee search successfully, but the node selects the next hop based on the node identifier(ID) in the logical spacing, which passing through different autonomy systems resulting inlong latency. Additionally, all of the location algorithms assume that all nodes in a P2P systemhave the same process capacity. However, the capacity of the nodes in the P2P network isextremely heterogeneous, which leads to load imbalance and reduces the efficiency of thelocation.Aiming at these problems above, this paper studies on resource location and loadbalancing in structured P2P and makes the following two contributions:1. In order to improve search performance of structured systems, this dissertation putsforward LChord according to the locality of reference (includes spatial and temporal locality)in network communication. Using XML technology, we constructs a file sharing system—PPFile based on LChord.Chord is a structured P2P of circle topology. As it has the neat structure and betterscalability, Chord is used widely. LChord, which is based on Chord, makes improvement inthe following two aspects:Firstly, LChord improves the finger selection according to the spatial locality and thusreduces search latency in each hop. Peers in Chord forward requests according to their fingerlist. As constructing the finger list, a peer in Chord selects the nearest one in the ID space asits finger and neglects the distance in physical network. While in LChord, a peer selects theone which is in the same autonomy system (AS) as its finger in the condition of keepingrouting correctness. As the AS of a peer seldom changes, it enhances the stability of the fingerlist and thus reduces system costs. This selection also embodies the temporal locality innetwork communication, i.e. keeping communication with the same node. Peers in LChordforward requests according to the improved finger list and thus acquire low latency.Secondly, LChord uses a location algorithm based on data relativity,so it has a shortersearch path. LChord creates an interest-based community for each node according to therelativity of share data. When initiating a request, the peer first searches in the community. Ifthe community can't match the request, the peer then forwards it according to the finger list.As peers always store files that they are interested in and may find data needed on the onesthat share common interests, LChord lessens the search scope and shortens search path. Whenit initiates a request, the peer checks its community and the community becomes the objectionthat it communicates with repeatedly, which shows the temporal locality of networkcommunication.Simulation results indicate that LChord has a lower latency and shorter search path thanChord.This paper, using XML technology, presents a P2P file sharing system based onLChord— PPFile, which supports exact-match lookup and fuzzy search simultaneously.2. Aimed at improving the degree of load imbalance in structured P2P, this dissertationpresents a low cost list-based load balancing algorithm (LCLLB). LCLLB mainly includes thefollowing two components:First, in condition of keep the fault tolerance, LCLLB uses a clustering virtual serverselection (CVSS) algorithm to reduce the costs of maintaining virtual servers. Traditionally, asvirtual servers are selected randomly in ID space, it results an excess cost. CVSS takes thecapacity heterogeneity of peers into consideration to allocate appropriate number of virtualservers for each peer and has these virtual servers clustered in a section in the ID space, so itreduces the system's extra cost.Second, LCLLB uses listed-based load balancing (LLB) algorithm to reduce thecommunication spending and keep reliable load balancing. LLB takes advantage of theperiodical communication among peers that keeps the topology of structured P2P to exchangeload information. This information is stored in heavy node list (HNL) or light node list (LNL)according to the node utilization. Nodes perform emergency load balancing or periodical loadbalancing according to their utilization to move virtual servers in HNL to peers in LNL. Thus,LLB establishes load balancing in the system.Simulation results show that LCLLB can provide reliable load balancing and low extracosting, which means better practicability.