In the last decade, MIMO spatial multiplexing has promised to increase data throughput by orders of magnitude. However, we are yet to enjoy such improvement in real-world environments, as achieving the capacity of the MIMO channels requires rich scattering and preclude effective MIMO spatial multiplexing. Furthermore, the gains are limited by scattering of the environment rather than MIMO antennas.

In this talk, I would present ScatterMIMO, which uses smart surface to increase the scattering in the environment, to provide MIMO spatial multiplexing gain. Specifically, the smart surface pairs up with a wireless transmitter device say an active AP, and re-radiates the same amount of power as any active access point (AP), thereby creating virtual passive APs. ScatterMIMO allows the smart surface to provide spatial multiplexing gain, which can be deployed at a very low cost. Furthermore, ScatterMIMO can provide signals to their clients with power comparable to real active APs, and can increase the coverage of an AP. Furthermore, we design algorithms to optimize ScatterMIMO’s smart surface for each client with minimal measurement overhead, provides a median throughput improvement of 2x over the active AP alone.

There are certain factors that render the cellular ecosystem complex and obfuscate the relationships between different networks and carriers, including the use of IP Exchange (IPX) providers that enable roaming partnerships or the presence of international carriers that sell connectivity services to vertical industries and connected machines across different countries. In this talk, I will try to shed light onto the structure of the cellular ecosystem and analyze the implications and exploitation of one of its fundamental features: international roaming. “Roam like Home” is the recent initiative (June 2017) of the European Commission (EC) to end the levy of extra charges when roaming within the European region. As a result, people are able to use data services more freely across Europe. However, the implications roaming solutions have on performance and user quality of experience received little examination. Even more, this infrastructure that mobile operators have established over the last 20 years for person-to-person communications is now being re-purposed to support the IoT expansion and M2M communication. Players across different markets -- from health to logistics -- require reliable connectivity and/or mobility around the world without needing to establish commercial relationships with every operator. We use real-world measurements to offer an in-depth characterization of roaming solutions and their implications in terms of performance or QoE for end-users traveling in Europe. We then focus the analysis on characterizing roaming for M2M/IoT communications. By combining the view from an operational M2M platform with that from an operational MNO in the UK, we shed light on the breadth of M2M/IoT platforms and the impact they have on visited MNOs.

Popular web services are often deployed across data centers in multiple regions in order to deliver low latency and high availability to users. In such global-scale deployments, service providers face several fundamental tradeoffs. For example, replicating data at all locations enables low latency reads, but at the expense of high storage and wide-area bandwidth costs. Similarly, a cloud provider is forced to choose between offering predictable bandwidth to tenants by statically reserving capacity versus maximizing the network's utilization by dynamically redistributing bandwidth among tenants.

In this talk, I will present some of the recent work in my group towards addressing the above-mentioned tradeoffs. In the context of geo-distributed storage, I will present Pando, our new approach for consensus across the wide- area network which does not merely improve upon the status quo, but achieves near-optimal latency versus cost tradeoffs. I will also discuss HEYP, our new architecture for global software-defined wide-area networks which simultaneously enables bandwidth predictability and high network utilization.

The focus of this work is on the benefits of using storage/computation resources at the edge to host Software as a Service (SaaS) instances like social networks, online shopping, navigation services etc. A key characteristic of current edge computing platforms is the pay-as-you-go flexibility via short term contracts. The client can thus optimize the cost incurred to edge resources by dynamically adjusting hosting decisions depending on request patterns. Unlike most of the existing literature on this topic, where the SaaS is either completely hosted at the edge or not hosted at all, we allow the service to be partially hosted at the edge. This is motivated by the growing popularity of the paradigm of designing services as a collection of independent microservices instead of a monolith.

In the setting where edge resources can be rented and the SaaS can be partially hosted at the edge, we first characterize the fundamental limit on the performance of any policy. We propose an online hosting policy which dynamically determines what fraction of the SaaS is hosted at the edge based on the request arrival process and the various costs involved in an online manner. We corroborate our analytical results by comparing the performance of the proposed policy and some natural alternatives like TTL via trace-driven simulations using real-data.

Joint work with Mohit Agarwala, Lakshmi Narayana, and Nikhil Karamchandani.

Internet Service Providers need to deploy and maintain many wireless sites in isolated or inaccessible terrain to provide Internet connectivity to rural communities. Addressing failures at such sites can be very expensive, both in identifying the fault, and also in the repair or rectification. Data monitoring can be useful, to spot anomalies and predict a fault (and possibly preempt it altogether), or to locate and isolate it quickly once it causes an issue for the network. There might be hundreds of variables to be monitored in principle, but only a few of significance for detecting faults. Here, in a case study involving a Wireless Internet Service Provider (WISP) in a rural area, we first illustrate a bottom-up approach to the identification of variables likely to be of use in an automatic anomaly detector. For the purpose of this study, the detector consists of an autoencoder neural network with weights optimized by machine learning (ML). We then show how the cause of an anomaly can be derived from indirect measurements, and use the model to learn relationships between certain variables.

Advances in mobile (cellular) networks have ushered in an era of abundant connectivity. However, the stationary and expensive nature of their deployment has limited their ability to provide true "ubiquitous" connectivity under the 5G vision - especially to areas where connectivity is sparing or nonexistent (e.g. rural areas), has been compromised (e.g. disasters), or demands are extreme (e.g. venues/hotspots).

The recent advances in un-manned aerial vehicle (UAVs) technology have the potential to change the landscape of wide-area wireless connectivity by bringing a new dimension - "mobility" to the cellular network infrastructure itself. By deploying base stations on each of the UAVs, service providers can now deploy and tear-down these cellular networks "in the sky" in an on-demand and flexible manner. This allows them to supplement static mobile networks in areas where additional connectivity is needed, or provide stand-alone connectivity in areas where existing mobile networks are either absent or compromised. However, realizing this vision of deploying heavy-weight cellular networks (e.g. LTE) on light-weight, resource-constrained platforms such as UAVs, faces several formidable challenges both in design and deployment. This is complicated by the complex nature of cellular networks that involve multiple interacting components - radio access network (RAN), evolved packet core (EPC) network and backhaul transport network.

In this talk, I will present our system "SkyLiTE"-- one of the first efforts to design and deploy an on-demand, end-to-end, multi-cell LTE network (on UAVs) that can self-configure itself in the sky. I will discuss how SkyLiTE brings together several advances in wireless networking research to re-architect the various components (RAN, core and backhaul) of a cellular network, thereby making it deployable on challenging UAV platforms in highly dynamic environments. The validation of SkyLiTE in real-world deployments shows the potential for contributing to a new era of mobile networks that can fly and make not just connectivity but also sensing ubiquitous.

In this talk, we consider a general version of opinion dynamics in discrete-time multi- agent networks when interactions among agents are either cooperative (trust) or antagonistic (mistrust). In particular, we study cluster consensus problems wherein agents of the network partition into more than two disjoint sets with non-identical opinions. Structurally balanced networks are well studied in classical opinion dynamics. In the first half of the talk, we will see its extension to cluster consensus problem and propose distributed iterative synchronous update rule to achieve cluster consensus over structurally balanced networks. We will see its extension to asynchronous and time-varying systems. In the second half of the talk, we show that structural balance network is a sufficient but not a necessary condition for achieving cluster consensus. We present sufficient conditions for the system stability and propose a control mechanism to achieve cluster consensus in structurally unbalanced networks.