An Efficient Cluster-Based Approach for Evaluating Vertical and Horizontal Scalability of Web Servers using Linear and Non-Linear Workloads

This paper attempts to address a very important question, i.e. whether a server should be scaled vertically or horizontally for linear and non-linear algorithms? Experiments were conducted with the aim to saturate web servers in a cluster while running a mixed workload of linear/super-linear algorithms while scaling it vertically (scale-up) and horizontally (scale-out), and comparing the increments achieved in the scalability i.e. Relative Capacity R, used a measure. The web servers were scaled vertically by increasing the number of enabled cores on a server node and scaled horizontally by increasing the number of server nodes involved in the experimental setup. The paper concludes by comparing the results of these experiments and examining the benefits of both these approaches.