Improving the performance of peer-to-peer systems
详细信息
- 作者：Sun ; Qixiang ; Ph.D.
- 学历：Ph.D.
- 年：2010
- 关键词：Peer-to-peer ; Rate limiting ; Link latency
- 导师：Garcia-Molina, Hector
- 毕业院校：Stanford University
- 专业：Computer science
- ISBN：9781109583656
- CBH：3395873
- Country：USA
- 语种：English
- FileSize：2902157
- Pages：201
文摘
Peer-to-peer (P2P) is an architecture where individual computers connect to each other to form a network. Unlike the centralized server-client model where everyone connects to the server, each peer is only connected to a few other peers in the network. When a peer needs information from the network, it relies on these directly connected neighboring peers to reach the rest of the network. For example, to search for a piece of data, a peer asks all its neighbors whether they have the desired data. If its neighbors do not have the data, then it asks the neighbors' neighbors and so on.
Without any dedicated central components, P2P provides three benefits: (1) P2P system is cheap to operate, (2) users retain full control over how they participate in the network, and (3) there is no single point of failure. However, there are many deficiencies in P2P as well. For instance, to find a piece of information, a peer may need to contact many other peers in the network, which is less efficient than contacting just one central server. Similarly, two directly connected peers may be on different continents with very high communication latency.
In this dissertation, we devise mechanisms and solutions to address some of the drawbacks. We make the following contributions. First, we give algorithms for rate limiting how quickly each peer in an unstructured P2P network is allowed to initiate new searches. This rate limiting guarantees that the network is used most efficiently. Second, we devise an incentive scheme to combat undesired peer behaviors, such as free-loading on the network without contributing anything in return. Third, we provide a mechanism to improve the average link latency in a P2P network. As a result, neighbors on the P2P network are also close in the physical network. Fourth, we examine various approaches to reduce the storage space required for tracking which peer has which data. Fifth, we propose a hybrid design that combines the benefits of unstructured and structured P2P systems. And lastly, we show how to build a P2P application in an ad-hoc wireless vehicle network.
Without any dedicated central components, P2P provides three benefits: (1) P2P system is cheap to operate, (2) users retain full control over how they participate in the network, and (3) there is no single point of failure. However, there are many deficiencies in P2P as well. For instance, to find a piece of information, a peer may need to contact many other peers in the network, which is less efficient than contacting just one central server. Similarly, two directly connected peers may be on different continents with very high communication latency.
In this dissertation, we devise mechanisms and solutions to address some of the drawbacks. We make the following contributions. First, we give algorithms for rate limiting how quickly each peer in an unstructured P2P network is allowed to initiate new searches. This rate limiting guarantees that the network is used most efficiently. Second, we devise an incentive scheme to combat undesired peer behaviors, such as free-loading on the network without contributing anything in return. Third, we provide a mechanism to improve the average link latency in a P2P network. As a result, neighbors on the P2P network are also close in the physical network. Fourth, we examine various approaches to reduce the storage space required for tracking which peer has which data. Fifth, we propose a hybrid design that combines the benefits of unstructured and structured P2P systems. And lastly, we show how to build a P2P application in an ad-hoc wireless vehicle network.