RP2: a high-performance data center network architecture using projective planes

Abstract

A data center network (DCN) plays a crucial role in data center communications and provides the infrastructure for cloud computing services and data-intensive applications in a cloud data center. Network performance is determined by DCN architecture, whose effective design critically depends on ensuring low cost, high availability, and high scalability. Currently, DCNs face undesirable trade-off between high availability, scalability and performance. In this paper, first, we propose a hybrid projective plane-based DCN architecture called the recursive projective plane (RP2). RP2 is underlain by the use of multi-port servers and mini-switches in DCN development and the application of a recursive method for scaling up a data center. It deals with the construction and properties of combinatorial designs with arrangements that satisfy the concepts of an efficient DCN. Then, we propose single-path and multi-path routing algorithms for RP2. Theoretical analysis indicates that RP2 provides a scalable architecture with a low diameter, wherein the maximum shortest distance between any pair of servers can be limited by level of extension. An empirical simulation that compares BCube, DCell, and RP2 indicates that under 0–20% server failure, RP2 exhibits efficient fault tolerance. Despite the higher cost and wiring complexity of RP2, it constitutes a high-performance and highly scalable DCN architecture.