Load-Balanced Of Lvs Scheduling Based On Adaptive Niche Genetic Algorithm

With the development of the Internet, the number of network users gradually increase. Thus, a server system that is able to cope with the mass concurrence accessing is needed. Nowadays, server cluster that accomplishes services base on paralleling servers is the most effective approach to solve the problem of server overloading. The server system is the equivalent of one super server for the client-side. Furthermore, the reliability of such system is high without the need of modifying the codes at the servers and clients.The existing scheduling algorithm of server cluster is very simple with relatively poor performance. In order to improve the performance of cluster system, we have investigated the algorithm and architecture of the Linux Virtual Server (LVS) in this paper. Firstly, the applications of genetic algorithm (GA) in scheduling works are studied. According to the different requirements of the load scheduling, the algorithm of adaptive niche genetic (ANGA) can be chosen for the server clusters scheduling. Moreover, we have proposed the relative CPU utilization using as the evaluation factor and applied it into the scheduling of load balancing. Secondly, the principle and the disadvantages of the LVS have been investigated. The architecture of regular server clusters has been improved by training the scheduling sequence based on decision maker. Furthermore, we also have designed and realized four modules, such as the scheduling algorithm, the communication unit between the load balancer and the decision maker, the communication unit between the decision maker and nodes, as well as the unit of modified IPVS kernel. Afterwards, the corresponding communication protocols and operating procedures are demonstrated in detail.Finally, we experimentally constructed a server cluster platform in the mode of LVS-NAT to measure the performance of the improved load balancing system. In the experiment, the stress-testing software of JMeter and the Wireshark are used to simulate mass concurrence and monitor the network data package, respectively. The results show that the average responding time of merely2-millisecond and the error rate of～0are achieved for the improved system, which is more stable compared with the conventional system.