Introduction to Peer-to-peer Technology

1.p2p Network Model

Peer-to-peer Network is a kind of distributed system structure with high expansibility, and its Peer-to-peer concept refers to that the physical nodes in the network have the same position logically, rather than the equivalence of processing ability. The Peer-to-peer technology which is represented by Napster software is the essence of directing the centralized management mode of Internet to the decentralized management mode, directing the content from the central single node to the edge of the network, so as to make full use of the processing ability and potential resources contained in many terminal nodes of Internet. Compared with the traditional centralized client/server (c/s) model, Peer-to-peer weakens the concept of the server, each node in the system no longer distinguishes between server and client role, each node can request service, also can provide services, between nodes can directly exchange resources and services without having to pass the server.

The most important feature of Peer-to-peer system is the direct sharing of resources between users, and its core technology is the location mechanism of distributed objects, which is the key to improve network scalability and solve the network bandwidth being swallowed. So far, Peer-to-peer Network has experienced three generations of different network models, various models have their advantages and disadvantages, and some of them are difficult to overcome the defects, so at present Peer-to-peer technology is far from mature stage, a variety of network structure can still coexist, and even show the form of mutual reference.

1.1 Set of directory-style structure

The centralized catalogue peer-to-peer structure is the first Peer-to-peer application model, because the central feature is also called the Non pure Peer-to-peer structure. Napster, which is used to share MP3 music files, is one of the most typical representatives (see Figure 1), whose user registration and file retrieval process is similar to the traditional C/s mode, except that all data is stored in each node rather than on the server. The query node establishes a direct connection based on information such as network traffic and latency, without having to go through a central server. This kind of network structure is very simple, but it shows the advantage and attraction of the information of Peer-to-peer system, and also reveals two unavoidable problems of peer-to-peer system in nature: legal copyright and resource waste.

1.2 Pure Peer-to-peer Network model

Pure Peer-to-peer mode is also known as the broadcast Peer-to-peer model. It cancels the centralized central server, where each user randomly accesses the network and forms a logically-covered network with its neighboring group of neighboring nodes through an end-to-end connection. The content query and content sharing between peers are transmitted directly through neighboring nodes, while each node also records the search trajectory to prevent the generation of search loops.

The Gnutella model is the most widely used pure Peer-to-peer unstructured topology (see Figure 2), which solves the problem of central network structure, scalability and fault tolerance, but the search algorithm in Gnutella network is in flood mode, Controlling the deluge of information consumes a lot of bandwidth and quickly creates network congestion and even network instability. At the same time, a node with poor local performance may cause the Gnutella network to be fragmented, resulting in poor availability of the entire network, which is more susceptible to spam and even malicious attacks by viruses.

1.3 Hybrid Network Model

Kazaa model is a typical representative of Peer-to-peer hybrid model (see Figure 3), it introduces the concept of Super node on the basis of pure peer-to-peer distributed model, and combines the advantages of centralized Peer-to-peer fast lookup and pure peer-to-peer centralization. The Kazaa model distinguishes the nodes according to their ability (computational ability, memory size, connection bandwidth, network retention time, etc.) as common nodes and search nodes (some of which are further divided into three types of nodes, whose thought is essentially the same). Where the search node forms an autonomous cluster between the common nodes adjacent to it, the cluster adopts Peer-to-peer mode based on centralized catalogue, and the whole Peer-to-peer network connects the search nodes through pure Peer-to-peer mode, and can even select the best performance node among each search node. Alternatively, a new optimal node is introduced as an index node to preserve the information of the search nodes available in the whole network, and it is responsible for maintaining the structure of the whole network.

As the common node of the file search first in the local cluster, only when the query results are not sufficient, and then through the search node for a limited flood. In this way, it is very effective to eliminate the negative effects such as network congestion and search delay caused by the flooding algorithm in pure peer-to-peer structure. At the same time, because the search nodes in each cluster monitor the behavior of all normal nodes, this also ensures that some malicious attacks can be controlled locally, and that the presence of super nodes can improve the load balance of the whole network to some extent.

In a word, the structure of hybrid Peer-to-peer network based on Super node has been improved greatly.

However, because of the fragility of the Super node itself, the nodes in the cluster can be isolated, so the local indexing method still has some limitations. This leads to the emergence of structured peer-to-peer network models.

1.4 Structured Network Model

The fundamental difference between the so-called structured and unstructured models is whether the neighbors maintained by each node can be organized in a global way to facilitate quick lookups. Structured Peer-to-peer mode is a kind of location service with pure distributed message passing mechanism and search based on keyword, the current mainstream method is to use distributed hash Table (DHT) technology, which is one of the best Peer-to-peer routing methods at present. As the DHT nodes do not need to maintain the information of the whole network, and only store the neighboring node information in the node, so less routing information can effectively reach the target node and eliminate the flood algorithm. This model can effectively reduce the number of node information sent, thus enhancing the scalability of peer-to-peer networks. At the same time, for redundancy and latency reasons, most DHT always replicates backup redundancy information on the node's virtual identity and the nodes closest to the keyword, which avoids the problem of single node failure.

Current representative research projects based on DHT include the can and tapestry projects at the University of California, Berkeley, the MIT Chord Project, the Iris project, and the pastry project of Microsoft Research. These systems generally assume that nodes have the same capabilities, which is more effective for smaller systems. However, this assumption is not appropriate for large-scale Internet deployments. At the same time, the topology maintenance and repair algorithm based on DHT is much more complicated than the unstructured system such as Gnutella model and Kazaa model, and even has a "detour" problem in the chord project. In fact, a large number of practical applications are based on unstructured topology and flood broadcasting, most of the Peer-to-peer systems using DHT are lack of real-world deployments in the Internet, and successful applications are still relatively rare.

2.p2p Network Application Mode

One of the main driving forces for the emergence and development of the Internet is resource sharing, and it is the demand for file exchange that leads directly to the rise of Peer-to-peer technology, which is one of the first and most successful applications of Peer-to-peer, and it is the Napster for such applications that makes people In server mode, we begin to recognize the influence of Peer-to-peer thinking on people's use of internet habits.

With the understanding of Peer-to-peer thinking and the development of technology, as a software architecture, Peer-to-peer can also be developed a wide range of application patterns, in addition to the initial file exchange, there are some distributed storage, depth search, distributed computing, personal instant messaging and collaborative work, and other novel applications. One of the most notable examples of this is the search for extraterrestrial civilizations based on distributed computing. SETI@home scientific experiments, each volunteer participant simply downloads and runs a similar screen saver, can contribute their idle computing power, participate in analyzing the wireless electromagnetic wave data of the Arecibo radio telescope and echo the data, As of December 2004, 5.28 million of volunteers have participated in the equivalent of 2.16 million years of CPU time, only one day of comprehensive computing is equivalent to 67.46Tflops operations. In addition, Peer-to-peer technology in mobile communications and intelligent Networks has also begun to show great application prospects as sun has extended its jxta protocol to hand-held terminals such as personal digital Assistants (PDA) and mobile phones, and allows people to block specific physical platforms for data sharing and file Exchange.

problems in 3.P2P network

The biggest advantage of Peer-to-peer is that it can provide reliable information inquiry, but from the social and legal sense, the vast majority of Peer-to-peer services will inevitably encounter intellectual property conflicts, but also may become some illegal content dissemination platform. At the same time, due to the lack of central supervision and the dynamic characteristics of free and equal, the self-organizing peer-to-peer network has many problems that are difficult to solve at the technical level.

In a sense, peer-to-peer networks and interpersonal networks have a certain similarity. Generally speaking, each Peer-to-peer network is a virtual organization formed by many participants according to the common interest, and there is a hypothetical mutual trust relationship between nodes, but as the scale of Peer-to-peer network expands, The dynamic characteristics of equality and freedom peculiar to these peer-to-peer nodes often contradict the trust collaboration model required by network services. The absence of the incentive to show more "greed", "complaints" and "fraud" selfish behavior, so peer-to-peer trust mechanism is actually very fragile, and this trust is difficult to reason between the nodes, resulting in a lack of overall trust, which directly affect the entire network stability and usability. In addition, compared to the traditional client/server model of the server can do active and passive defense, due to the lack of Peer-to-peer node security measures and Peer-to-peer protocol lacks the necessary authentication mechanism and computer operating system security vulnerabilities, security issues in the Peer-to-peer network more serious, which will directly affect peer-to-peer large-scale commercial. In addition, the Peer-to-peer network node itself is often the difference in computing power of heterogeneous nodes, each node is given the same responsibility without considering its computing power and network bandwidth, poor local performance will lead to the deterioration of the overall network performance, It is difficult to achieve optimized resource management and load balancing in this heterogeneous node environment. At the same time, because of the randomness of users ' joining out of Peer-to-peer network, the user obtains the uncertainty of the target file, which leads to many unnecessary file downloads, and causes the abuse of the large bandwidth resources. In particular, most peer-to-peer users prefer to send audio, video these larger media files, which will make the problem of bandwidth wasting more prominent, especially in China, a large number of users or dial-up users, the narrow bandwidth has become a peer-to-peer application insurmountable obstacles.