Study Of Anycast Routing Algorithm Base On Dynamic Load Balancing

Various types of network application have emerged in our life with the booming of computer technology and network technology, meeting and arousing people's demand for network services, all that rise the volume of network service demand exponentially. Yet it is due to growth of network service demand, leading to network service capacity comparatively shortage. In order to enhance the availability of service, Internet Service Provider introducing replication server technology, they distribute multiple servers with the same content at different network position, and users get the nearest service they want, which improves the efficiency of information flowing in Internet. However, the consequent problems are that, how to guide the user to the"nearest"server, at the same time balancing the load among replication servers dynamically. Anycast routing technology emerges to solve these problems. The sharply increasing of the service traffic volume in the network puts tremendous pressure on the network links, and then leads to the congestion on network links. Though this nervous network status does not threaten the traditional non-real-time network application too much, it seriously affects quality of the real-time service, and reduces the satisfaction of users for network services, then restricts and affects long-term development and benefit of Internet Service Provider. In today's dynamic and complex network environment, how to guide user requests to the position of best replication server, provide"better"and"faster"network service to users, at the same time balance load both among the replication servers and network links, and rapidly do reaction to the change of network status. Dynamic load balancing routing technology has become the important problem in current researches.Dynamic load balancing routing technology includes two key function modules. First it should select the best server and compute a set of path corresponding. And then it should split the traffic on the candidate paths dynamically and reasonably. Hence, improvement for both anycast routing strategy and traffic split become hot issues in today's research. Due to the complexity of the network applications and the users'rising demand of network service quality. The QoS routing problem in current network is generally multi-constraint problem. When it comes to constraint composed with at least two addition or multiplication rules, routing is NP complete problem. Recently, scholars generally use heuristic algorithm to solve anycast routing problem with multiple QoS constraints, and one has vividly applied ant colony algorithm in the anycast routing problem and get good performance. Nevertheless due to the premature convergence property of the algorithm, the path ultimately can't converge around the global optimal and greatly consume the computation time. At the same time path convergence too much focused on some partial links, causes load imbalance that partial links are much too heavy while other links are free. Either the imbalance happens on the link or the anycast server in the network will make poor use of network resource, in addition to this, heavy load on both link and server could introducing packet losing, large delay, and so on, which seriously affects user's accessquality. View of the above problems, this paper proposes an anycast routing algorithm based on dynamic load balancing, there are main works as follows:1.Anycast routing based on dynamic load balancing should select a best anycast server and compute a set of path corresponding. To solve overwhelmingly premature convergence problem of the heuristic algorithm, this paper proposes that inserting the mutation operation of differential evolution algorithm into ant colony algorithm at the needed time. Making use of characteristic of differential evolution algorithm that: few controlled parameter, strong robust, easily combined with other algorithm and powerful global optimizing ability. It can expand the search space of ant colony algorithm, and make the anycast paths uniformly distributed. This strategy proposed in this paper does not retain an optimal path for data transmission, but calculates a set of better candidate paths. The strategy uses multipath parallel transmission technology to split the traffic onto multipath to achieve load balancing, and improves the reliability of service and throughput of the network.2.As to the problem that how to reasonably split the traffic onto the paths computed through the algorithm above. This paper analyses and does a fine tradeoff between the two classical splitting granularity, and improves the method of FLARE. When FLARE confronts with that time interval of successive packet is small while the difference between the maximum and minimum path latencies is larger, then the load is not in balance. This paper makes use of difference among multiple path latencies and designs a Improved FLARE (I-FLARE), set the splitting granularity smaller, forwards current packet onto the lightest load candidate path that meet the latency requirement. All that helps achieve to the goal of dynamic load balancing without packet reordering.Finally, this paper uses NS-2 network simulator to simulate the two key function modules of anycast routing algorithm based on dynamic load balancing respectively. The analysis of experimental results proves that the hybrid strategy increases the population diversity when anycast routing algorithm falling into the premature convergence, then help the algorithm escape from the premature convergence and enhance global search capability. I-FLARE proposed in this paper could ensure packets belonging to the same flow arrive at the object node without reordering, at the same time balancing load among the network link dynamically. The improved strategy greatly improves network performance.