Hardware Sizing for Software Application

Abstract

Hardware sizing is an approximation of the hardware resources required to support a software implementation. Just like any theoretical model, hardware sizing model is an approximation of the reality. Depending on the infrastructure needs, workload requirements, performance data and turn around time for sizing, the study (Sizing or Capacity Planning) can be approached differently.

The most common method is to enter all the workload-related parameters into a modeling tool that is built using the results of workload simulation on different hardware. The hardware and software requirements are determined by the mathematical model underlying the tool. Without performing a test on the actual hardware environment to be used, no sizing can be 100% accurate. However, in real-life there is a need to predict the capacity when budgeting hardware, assessing technical risk, validating technical architecture, sizing packaged applications, predicting production system capacity requirements, and calculating the cost of the project. These scenarios call for a quick way to estimate the hardware requirements. When dealing with prospects, there is a need to come up with credible and accurate sizing estimates without spending a lot of time.

One of the challenges faced by Kronos is the amount of effort and time spent in hardware sizing for prospective customers. Typically, a survey process collects the workload related parameters and feeds the sizing tool, which uses the performance model based on benchmark test results to produce the hardware recommendations. Although this process works great for customers, it is a time consuming activity due to the collection and validation of large number of independent variables involved in the current sizing model.

This project makes an attempt to delve into alternate methods for producing quick sizing. By combining the empirical data collected from various production systems and simple statistical technique, relationship between sizing factors and CPU rating can be established. This can be used to create a simple model to produce a quick, easy and credible recommendation when sizing new customers.