Reducing the Cost of Operating a Datacenter Network

Datacenters are a signi cant capital expense formany enterprises. Yet, they aredifficult tomanage and are hard to design andmaintain. e initial design of a datacenter network tends to follow vendor guidelines, but subsequent upgrades and expansions to it are mostly ad hoc, with equipment being upgraded piecemeal a er its amortization period runs out and equipment acquisition is tied to budget cycles rather than changes in workload. ese networks are also bri le and in exible. ey tend to be manually managed, and cannot perform dynamic traffic engineering. e high-level goal of this dissertation is to reduce the total cost of owning a datacenter by improving its network. To achieve this, wemake the following contributions. First, we develop an automated, theoretically well-founded approach to planning cost-effective datacenter upgrades and expansions. Second, we propose a scalable traffic management framework for datacenter networks. Together, we show that these contributions can signi cantly reduce the cost of operating a datacenter network. To design cost-effective network topologies, especially as the network expands over time, updated equipment must coexist with legacy equipment, which makes the network heterogeneous. However, heterogeneous high-performance network designs are not well understood. Our rst step, therefore, is to develop the theory of heterogeneous Clos topologies. Using our theory, we propose an optimization framework, calledLEGUP,whichdesigns a heterogeneousClos network to implement in anewor legacy datacenter. Although effective, LEGUP imposes a certain amount of structure on the network. To deal with situations when this is infeasible, our second contribution is a framework, called REWIRE, which using optimization to design unstructuredDCN topologies.Our results indicate that these unstructured topologies have up to more bisection bandwidth than a fat-tree for the same dollar cost. Our third contribution is two frameworks for datacenter network traffic engineering. Because of the multiplicity of end-to-end paths in DCN fabrics, such as Clos networks and the topologies designed by REWIRE, careful traffic engineering is needed to maximize throughput. is requires timely detection of elephant ows— ows that carry large amount of data—and management of those ows. Previously proposed approaches incur high monitoring overheads, consume signi cant switch resources, or have long detection times. We make two proposals for elephant ow detection. First, in the Mahout framework, we suggest that such ows be detected by observing the end hosts’ socket buffers, which provide efficient visibility of owbehavior. Second, in theDevoFlow framework, we add efficient stats-collection mechanisms to network switches. Using simulations and experiments, we show that these frameworks reduce traffic engineering overheads by at least an order of magnitude while still providing near-optimal performance.