Reducing the Cost of Operating a Datacenter Network

Опубликовано 12 августа 2016, 0:29

Datacenters are a significant capital expense for many enterprises. Yet, they are difficult to manage and are hard to design and maintain. The initial design of a data center network tends to follow vendor guidelines, but then upgrades and expansions to it are mostly ad hoc, with equipment being upgraded piecemeal after its amortization period runs out and equipment acquisition tied to budget cycles rather than changes in workload. These networks are also brittle and inflexible. They tend to be manually managed, and cannot perform dynamic traffic engineering. The high-level goal of this dissertation is to reduce the total cost of owning a datacenter by improving its network. To be flexible, a datacenter network should be able to incorporate heterogeneous switch types. Our first step, therefore, is to develop the theory of heterogeneous Clos topologies. Using this, we propose an algorithm LEGUP which designs a heterogeneous Clos network for a legacy data center. Although effective, LEGUP imposes a certain amount of structure on the data center network. To deal with situations when this is infeasible, our second contribution is a framework, called REWIRE, to design unstructured networks for existing data centers. Our results indicate that unstructured topologies have 100-500\ more bisection bandwidth than regular topologies like a fat-tree for the same dollar cost. Finally, to simplify the management of these networks, our third contribution is a framework for scalable datacenter flow management. This enables cost-reducing and innovative network management solutions for the datacenter. As a use case, we evaluate the effectiveness of our approach at reducing the overheads of traffic engineering in the datacenter.