Load Balancer Mechanism using Optimal Parameter based on Calculus

This research proposes improvement of load balancer mechanism using technique that agent software preinstalled on servers has to read server resource information and send the information to load balancer. This technique provides reliable and up-to-date information to the load balancer because if the information, e.g. active connections, is estimated or counted by load balancer, it may have some errors or be not match with the real information. Also, putting weights of each server to the load balancer by an administrator may have unintentionally bias. With this technique, ‘plan making’ for dispatching incoming requests to servers will operate efficiently. Implementing the agent for reading and sending resource information to the server can be done easily. Similarly, an algorithm to create the plan making function is not hard because it is similar to other existing load balancing algorithms. Therefore, the main problem for this research is that how much of time interval between each plan making should be appropriate. This is because too wide of time interval can make slow reaction of system. Consequently, the overall efficiency will be dropped. Likewise, too narrow of time interval can bring the drop of overall efficiency as well. This is because the load balancer has to use almost of all its CPU time to make a dispatching plan. Thus, investigating to find the best value of time interval is the main focus of this research. Highlight of the proposed method is discovering the optimal time interval ‘w’ which can bring the highest overall efficiency ‘P’ to the network. We firstly present the function of overall efficiency in the term $P =$ ${f(w)}$ and then differentiate the function to obtain its derivative function. Consequently, we can discover the optimal ‘w’ at the point that ‘P’ can reach to its maximum.