Performance evaluation of statefull load balancing in predicted time intervals and CPU load

摘要

Internet service providers are increasingly facing the growing demand for higher bandwidth to satisfy users demand. In the condition of overloaded, the web servers show delays in offering services until reaching total unavailability. There exist several approaches to deal with this problem, the core of them use the configuration of a web server in cluster, but offering the service as a single entity. The ways to distribute the load on the web servers' cluster vary from the simplest to the most complicated ones, which analyze the state of the servers and optimize the balance. In this paper is presented a new evaluation method of the servers load. The algorithm calculates the CPU load based on the amount of traffic exchanged from each server and the number of active connections, weighting more the later factor. The servers under a specified safety level of the CPU usage, undergo a weighted evaluation process to determine the least loaded one, at which the clients 'requests will be sent. The NS2 simulation results show a significant reduction in balancer work, while increasing its performance on traffic distribution.