Data Driven Network Design for Cloud Services Based on Historic Utilization

Abstract

In recent years we have seen a shift from traditional networking in enterprises with Data Center centric architectures moving to cloud services. Companies are moving away from private networking technologies like MPLS as they migrate their application workloads to the cloud. With these migrations, network architects must struggle with how to design and build new network infrastructure to support the cloud for all their end users including office workers, remote workers, and home office workers. The main goal for network design is to maximize availability and performance and minimize cost. However, network architects and network engineers tend to over provision networks by sizing the bandwidth for worst case scenarios wasting millions of dollars per year. This thesis will analyze traditional network utilization data from twenty-five of the Fortune 500 companies in the United States and determine the most efficient bandwidth to support cloud services from providers like Amazon, Microsoft, Google, and others. The analysis of real-world data and the resulting proposed scaling factor is an original contribution from this study.