Despite the many proposals for data center network (DCN) architectures, designing a DCN remains challenging. DCN design is especially difficult when expanding an existing network, because traditional DCN design places strict constraints on the topology (e.g., a fat-tree). Recent advances in routing protocols allow data center servers to fully utilize arbitrary networks, so there is no need to require restricted, regular topologies in the data center. Therefore, we propose a data center network design framework, that we call REWIRE, to design networks using an optimization algorithm. Our algorithm finds a network with maximal bisection bandwidth and minimal end-to-end latency while meeting user-defined constraints and accurately modeling the predicted cost of the network. We evaluate REWIRE on a wide range of inputs and find that it significantly outperforms previous solutions-its network designs have up to 100-500% more bisection bandwidth and less end-to-end network latency than equivalent-cost DCNs built with best practices.

Large-scale Internet applications, such as content distribution networks, are deployed across multiple datacenters and consume massive amounts of electricity. To provide uniformly low access latencies, these datacenters are geographically distributed and the deployment size at each location reflects the regional demand for the application. Consequently, an application's environmental impact can vary significantly depending on the geographical distribution of end-users, as electricity cost and carbon footprint per watt is location specific. In this paper, we describe FORTE: Flow Optimization based framework for request-Routing and Traffic Engineering. FORTE dynamically controls the fraction of user traffic directed to each datacenter in response to changes in both request workload and carbon footprint. It allows an operator to navigate the three-way tradeoff between access latency, carbon footprint, and electricity costs and to determine an optimal datacenter upgrade plan in response to increases in traffic load. We use FORTE to show that carbon taxes or credits are impractical in incentivizing carbon output reduction by providers of large-scale Internet applications. However, they can reduce carbon emissions by 10% without increasing the mean latency nor the electricity bill.

Datacenters need high-bandwidth interconnection fabrics. Several researchers have proposed highly-redundant topologies with multiple paths between pairs of end hosts for datacenter networks. However, traffic management is necessary to effectively utilize the bisection bandwidth provided by these topologies. This requires timely detection of elephant flows-flows that carry large amount of data-and managing those flows. Previously proposed approaches incur high monitoring overheads, consume significant switch resources, and/or have long detection times. We propose, instead, to detect elephant flows at the end hosts. We do this by observing the end hosts's socket buffers, which provide better, more efficient visibility of flow behavior. We present Mahout, a low-overhead yet effective traffic management system that follows OpenFlow-like central controller approach for network management but augments the design with our novel end host mechanism. Once an elephant flow is detected, an end host signals the network controller using in-band signaling with low overheads. Through analytical evaluation and experiments, we demonstrate the benefits of Mahout over previous solutions.

OpenFlow is a great concept, but its original design imposes excessive overheads. It can simplify network and traffic management in enterprise and data center environments, because it enables flow-level control over Ethernet switching and provides global visibility of the flows in the network. However, such fine-grained control and visibility comes with costs: the switch-implementation costs of involving the switch's control-plane too often and the distributed-system costs of involving the OpenFlow controller too frequently, both on flow setups and especially for statistics-gathering.

Fundamental limitations of traditional data center network architectures have led to the development of architectures that provide enormous bisection bandwidth for up to hundreds of thousands of servers. Because these architectures rely on homogeneous switches, implementing one in a legacy data center usually requires replacing most existing switches. Such forklift upgrades are typically prohibitively expensive; instead, a data center manager should be able to selectively add switches to boost bisection bandwidth. Doing so adds heterogeneity to the network's switches and heterogeneous high-performance interconnection topologies are not well understood. Therefore, we develop the theory of heterogeneous Clos networks. We show that our construction needs only as much link capacity as the classic Clos network to route the same traffic matrices and this bound is the optimal. Placing additional equipment in a highly constrained data center is challenging in practice, however. We propose LEGUP to design the topology and physical arrangement of such network upgrades or expansions. Compared to current solutions, we show that LEGUP finds network upgrades with more bisection bandwidth for half the cost. And when expanding a data center iteratively, LEGUP's network has 265% more bisection bandwidth than an iteratively upgraded fat-tree.

The OpenFlow framework enables flow-level control over Ethernet switching, as well as centralized visibility of the flows in the network. OpenFlow's coupling of these features comes with costs, however: the distributed-system costs of involving the OpenFlow controller on flow setups, and the switch-implementation costs of involving the switch's control plane too often.

Valiant load balancing (VLB), also called two-stage load balancing, is gaining popularity as a routing scheme that can serve arbitrary traffic matrices. To date, VLB network design is well understood on a logical full-mesh topology, where VLB is optimal even when nodes can fail. In this paper, we address the design and capacity provisioning of arbitrary VLB network topologies. First, we introduce an algorithm to determine if VLB can serve all traffic matrices when a fixed number of arbitrary links fail, and we show how to find a min-cost expansion of the network - via link upgrades and installs - so that it is resilient to these failures. Additionally, we propose a method to design a new VLB network under the fixed-charge network design cost model. Finally, we prove that VLB is no longer optimal on unrestricted topologies, and can require more capacity than shortest path routing to serve all traffic matrices on some topologies. These results rely on a novel theorem that characterizes the capacity VLB requires of links crossing each cut, i.e., a partition, of the network's nodes.

Routing policies play an essential role in how traffic is forwarded across the Internet. The network would not be commercially viable without these routing policies, but policies also introduce inefficiencies and fail to fully exploit the underlying network topology. Our work assumes routes are selected according to some policy such as a valley-free routing policy. However, we apply policy at an aggregate traffic level and don't require individual packets to follow paths that match the policy. Our approach never reduces, and usually increases, the connectivity and capacity of the network, and does not infringe on the underlying motivations that led to the routing policy. By adopting this approach, we also provide polynomial algorithms for otherwise NP-hard problems, including finding maximum policy-observing routing capacity between two sets of ASes, minimizing cuts separating all policy-observing paths between two sets of ASes, and maximizing sets of edge-or vertex-disjoint policy-observing paths.