Advancing Data Center Networks: A Focus on Energy and Cost Efficiency

Abstract

Data centers serve as the backbone for cloud computing, enterprise services, and infrastructure-based offerings. One area of ongoing research in data center networking focuses on innovating new topologies for large-scale node connectivity. These topologies must incorporate fault-tolerant and efficient routing algorithms. Consequently, the data center network topology must dynamically adapt to ever-changing application requirements. While traditional topology designs often emphasize scalability, they are typically limited by the necessity for dedicated switches to manage server connections. The development of software technologies that distinguish server and switch roles offers a unique opportunity to reconsider design priorities, paving the way for a more balanced assessment of scalability, energy efficiency, and infrastructure costs. Moreover, certain network topologies fail to be cost-effective due to their structural intricacies, often requiring far more node connections than those that are practically necessary. To address these challenges, we introduce VacoNet: a new flexible data center network topology that organizes nodes into structurally similar clusters, interconnected by a novel physical structure algorithm. Boasting high bisection bandwidth, VacoNet delivers robust network capacity, even when encountering bottlenecks. Furthermore, to connect a given set of nodes, VacoNet uses a minimal number of cables and switches, thereby drastically reducing both infrastructure costs and energy consumption. Simulation results show that VacoNet can reduce error rates by 20% and slash infrastructure costs by 70% compared to existing solutions. Additionally, it performs tasks 30% faster, underscoring its superior performance.