|
چکیده
|
In order to meet the insatiable demands for cloud-based services, today’s data centers may have hundreds of thousands of servers. During the past decade, many topologies have been proposed to interconnect a large number of servers in a data center network. However, existing topologies face several challenges including the rigid structure, high network cost, and packet relay overheads on servers. In this paper, we propose a novel switch-centric network topology, called DCCube, which offers many desirable properties for data center networking. DCCube is a flexible structure because, given the switches degree and the network diameter, it is possible to construct different structures with different number of connected servers, bisection bandwidth and network cost. Furthermore, DCCube can accommodate a large number of servers with high bisection bandwidth, low cost, small diameter and desirable path diversity. By comparing with existing data center network topologies, we show that DCCube achieves both high performance and low network cost. We also design both minimal and non-minimal routing algorithms for the proposed topology. Simulation results reveal that non-minimal routing is preferred for unbalanced traffic patterns.
|