UC San Diego

Data center architectures are gaining more and more interest nowadays due to the emerging need for hosting tens of thousands of workstations with significant aggregate bandwidth requirements. The underlying network architecture of a data center typically consists of a tree of routing and switching elements that count to thousands of components for the large-scale data centers that are deployed today. Recent research activity has been focusing on many of the challenges that one is faced with on building such large in scale data centers, such as scalability issues, dealing with the possibly inherent non -uniformity of the bandwidth among data center nodes, minimizing the actual cost of building and deploying these data centers, reducing the cabling complexity involved with interconnecting the routing and switching elements that a data center is composed of and coming up with efficient forwarding and routing mechanisms to be used within the data center. In this context, many solutions to efficient forwarding and routing have been proposed recently that take advantage of a switch' s relevant location in the global topology. This study focuses on a specific challenge regarding data center architectures, which is that of providing a plug and play functionality to a data center's switches so that no such switch will require any kind of human intervention or physical device configuration for figuring out its location in the underlying network architecture. We present a scalable, distributed and efficient Location Discovery Protocol for data center network topologies (LDP). Our solution is generic, allowing for any kind of multi-rooted tree architecture to be used when interconnecting the switching elements that constitute the core of a data center network