Recently, Multipath TCP (MPTCP) has been proposed as an alternative transport approach for datacenter networks. MPTCP provides the ability to split a flow into multiple paths thus providing better performance and resilience to failures. Usually, MPTCP is combined with flow-based Equal-Cost Multi-Path Routing (ECMP), which uses random hashing to split the MPTCP subflows over different paths. However, random hashing can be suboptimal as distinct subflows may end up using the same paths, while other available paths remain unutilized. In this paper, we explore an MPTCP-aware SDN controller that facilitates an alternative routing mechanism for the MPTCP subflows. The controller uses packet inspection to provide deterministic subflow assignment to paths. Using the controller, we show that MPTCP can deliver significantly improved performance when connections are not limited by the access links of hosts. To lessen the effect of throughput limitation due to access links, we also investigate the usage of multiple interfaces at the hosts. We demonstrate, using our modification of the MPTCP Linux Kernel, that using multiple subflows per pair of IP addresses can yield improved performance in multi-interface settings.

Recent years have shown a promising progress in understanding geometric underpinnings behind the structure, function, and dynamics of many complex networks in nature and society. However, these promises cannot be readily fulfilled and lead to important practical applications, without a simple, reliable, and fast network mapping method to infer the latent geometric coordinates of nodes in a real network. Here, we present HyperMap, a simple method to map a given real network to its hyperbolic space. The method utilizes a recent geometric theory of complex networks modeled as random geometric graphs in hyperbolic spaces. The method replays the network's geometric growth, estimating at each time-step the hyperbolic coordinates of new nodes in a growing network by maximizing the likelihood of the network snapshot in the model. We apply HyperMap to the Autonomous Systems (AS) Internet and find that: 1) the method produces meaningful results, identifying soft communities of ASs belonging to the same geographic region; 2) the method has a remarkable predictive power: Using the resulting map, we can predict missing links in the Internet with high precision, outperforming popular existing methods; and 3) the resulting map is highly navigable, meaning that a vast majority of greedy geometric routing paths are successful and low-stretch. Even though the method is not without limitations, and is open for improvement, it occupies a unique attractive position in the space of tradeoffs between simplicity, accuracy, and computational complexity.

We show that complex (scale-free) network topologies naturally emerge from hyperbolic metric spaces. Hyperbolic geometry facilitates maximally efficient greedy forwarding in these networks. Greedy forwarding is topology-oblivious. Nevertheless, greedy packets find their destinations with 100% probability following almost optimal shortest paths. This remarkable efficiency sustains even in highly dynamic networks. Our findings suggest that forwarding information through complex networks, such as the Internet, is possible without the overhead of existing routing protocols, and may also find practical applications in overlay networks for tasks such as application-level routing, information sharing, and data distribution.

Experimentation with mobile ad hoc network testbeds is preferred to simulations for performing high fidelity testing. But, at the same time, realistic experimentation with large-scale network testbeds is more difficult, time-consuming, and expensive. To side-step some of these problems several researchers, e.g. P. De et al. (2005), V. Naik et al. (2006), have recently suggested experimentation on scaled-down replicas and have managed to downscale some specific indoor network realizations with no mobility. However, the question of whether a scaled-down replica can reproduce the behavior of an arbitrary large-scale mobile ad hoc network, in a timely manner, and under realistic outdoor conditions, remains an interesting open problem. In this work we investigate ways of constructing suitably scaled-down replicas, both in space and time, that can predict the performance of large-scale mobile ad hoc networks with high accuracy. We consider both large- and small-scale fading effects. Further, we present necessary and sufficient conditions for the scaling to be possible, and identify some of the issues that may arise in practice. Finally, we argue that it is not possible to build arbitrarily smaller replicas, and that the factor by which one scales down the original network (in space) depends on the carrier frequency.

The Internet is a large, heterogeneous system operating at very high speeds and consisting of a large number of users. Researchers use a suite of tools and techniques in order to understand the performance of complex networks like the Internet: measurements, simulations, and deployments on small to medium-scale testbeds. This work considers a novel addition to this suite: a class of methods to scale down the topology of the Internet that enables researchers to create and observe a smaller replica, and extrapolate its performance to the expected performance of the larger Internet. This is complementary to the work of Psounis, 2003, where the authors presented a way to scale down the Internet in time, by creating a slower replica of the original system. The key insight that we leverage in this work is that only the congested links along the path of each flow introduce sizable queueing delays and dependencies among flows. Hence, one might hope that the network properties can be captured by a topology that consists of the congested links only. Using extensive simulations with transmission control protocol (TCP) traffic and theoretical analysis, we show that it is possible to achieve this kind of performance scaling even on topologies the size of the CENIC backbone (that provides Internet access to higher education institutions in California). We also show that simulating a scaled topology can be up to two orders of magnitude faster than simulating the original topology