Research On The Resilience Strategies Of Churn Of The Structured P2P Networks

P2P technology, as one of the most popular network computing technologies during the frist decade of the 21 st century, attracts the attention of academia, industry and common users, and many extensive and deep researches have taken place. However, there are still some problems affecting the P2 P application waiting to be resolved. Churn which means the collective effect created by the arrival and departure of thousands(or millions) of peers, is one particular aspect to be better solved. The influence of churn to the network is comprehensive, possibly causing problems like the inconsistency of topology, the performance reduction of network transmission and routing, the failure of network connection as well as the router. It is of high academic value and application value to present a brand new solution to the churn problem in P2 P network in a bid to improve stability.As for P2 P network, one of its main function is to publish and share data in the netwok, making it possible for interested users to find and get the data. The data availability directly affects the spreading and application of the P2 P network. However, churn can import many problems like data lost, data migration and data access latency increase. The replica technology is one of the main technologies to improve the data availability, but it is not be well applied for churn, which would lead to the replica node change at any time, causing problems like replica lost, replica migration and the like to happen. These would increase the maintenance overhead. So how to decrease the replica maintenance overhead and reduce the number of replica migration is the problem worthy of great concern in the replica maintenance to deal with churn.When the Leaf Set size is large, the problem the replica being concentrated is easy to happen. So thinking besides of reducing the replica migration times, we also need to think about to avoid the replica distribute imbalance. So under the churn situation, when the Leaf Set size is large, How to avoid the the replica distribution concentrated is the problem worthy of greate concern.If we want the P2 P network to run effectively, we must ensure that most of the nodes in the routing table are online. But churn can make the information of the routing table out of date, which means that when connecting to the nodes seemingly online the connection failure happens. In the churn situation, how to reduce the number of failure nodes in the routing table and how to ensure that the nodes in the routing table are almost the online nodes, are the questions what we need to consider when dealing with churn in maintenance of the routing table.One of the basic function of the P2 P network is the routing find function. But under the churn sitruation the next chosen hop is often failure when routing, which will cause the route failure. But the cost of re-connecting after failure is 3-5 times of 第 iv 页 ordinary connecting. For that, we need to think over a good strategy to select a stable node to send when sending routing information. But how to select a stable node to send the message is what we need to consider to deal with churn.This dissertation proposes several churn resilience strategies in the replica maintenance, routing table maintenance and routing selection in the structured P2 P networks. The main contribution of this dissertation is as follows:(1) In order to solve the problem that the replica maintenance overhead is high and need to be migrated many times in structured P2 P networks, this dissertation proposes an Age-based Replication Maintenance Strategy(ARMS) which can reduce the replica migration times. This strategy uses the local clock as reference, and appoints the time when the node receives the message of new node arriving to the network to be the new node birth time at this node. Then it sorts the Leaf Set nodes by the birth time. When choosing a neighbor node to store the replica, it selects the oldest non-replica node to be the replica storing place. The analysis and experiments show that, the strategy can select the stable node to store replica with high probability, and also can reduce the selection times before selecting the stable node compared with the random selection strategy.(2) When the leaf set size is large(for example L>24), the replication maintenance strategy should consider not only the replica stability, but also the replica distribute imbalance. The dissertation presents a Random-based Replication Load Balance Strategy(RRLBS). In this strategy, when choosing the replica place, the strategy first chooses ‘s’ nodes in the leaf set randomly to compose a candidate node set, and then selects the oldest non-replica node to be the replica storing place. The analysis and experiments show that, the strategy not only reduce the replica migration times, but also avoid the replica distribute imbalance.(3) For the problem that the routing table has many seemingly online but offline nodes, this dissertation presents the popularity-based routing table maintenance strategy. The strategy based on the idea that the node which is more popular will be more complimentary. In thinking over on the aspect of stability, these nodes are stable. In this strategy, each node has a list which stores all the nodes linking to this node, and the size of the list is named the popularity rate. When the node sends the message, it sends the popularity rate together. When the node receives the message, it computes which k-bucket is the source node of the message located in. If the bucket is not full, the message source node is added to the end of the k-bucket directly; if the k-bucket is full, the strategy compares the popularity rate of the message source node with the least popularity rate of the k-bucket. If the popularity rate of message source node is large, the k-bucket will deletes the node which has the least popularity rate, and adds the message source node to this k-bucket. Otherwise, nothing would be done. The analysis and experimentations show that, the strategy can reduce the number of offline node in the routing table.(4) In order to reduce the possibility of sending message to the offline node when routing, this dissertation presents the proportion-based routing find strategy. We first analyze the node stability of the network, and get the conclusion that the number of stable node in the network is proportioned, and distributes uniformly in the network. So when routing, the strategy selects ‘s’ nodes according to the distance to form a candidate node set, and adds the corresponding proportion factor ‘p’ based on the monitoring results, so selects ‘s*p’ nodes from the ‘s’ candidate node set according to the popularity. The experiment results show that, the strategy can select the stable nodes by high probability to send message, and reduce the message number of routing failure.