The Billion Dollar cloud: Architectural patterns for hyperscale cost optimization

Abstract

In this paper, we will explore the patterns that are crucial when designing data center infrastructure at hyperscale to minimize costs and enhance effectiveness in terms of latency, consistency, availability, and reliability of the deployed infrastructure. Findings indicate that up to 90% of network communication costs could be saved by dynamically provisioning on-demand compute resources. Inspired by fluid dynamics principles, server temperatures were observed to be up to 30% lower, resulting in higher power savings. The patterns to maximize the number of virtual machines on a physical server optimize performance by an average of 34.41%. The dynamic utilization of old and new hardware configurations results in a significant 40% reduction in hardware replacement costs and improves reliability. Predictive state monitoring and compute resource provisioning lead to over 20% reduction in power consumption. Operating at internet scale, this hybrid setup requires billions of dollars in operational costs for maintaining cloud and physical infrastructure. This paper describes architectural patterns that contribute to overall cost reduction when applied to different layers in cloud or on-premises hyperscale infrastructure.