The initial design of the Internet and its protocols did not impose limitations to performance as its purpose intended to use and share expensive computing resources. Modern networks consists of different communication models that expand the capabilities of how resources are exchanged with one another. Wireless sensor networks (WSN) and the Internet of Things (IoT) are examples that have become dominant network platforms for data acquisition, allowing service providers and consumers to collect and exchange data with sensors and devices embedded throughout the physical world. However, these types of networks suffer from the limitations of lossy communication media and low-powered devices. Additional drawbacks that these networks still face are that conventional design methods intended for traditional networks are still being used in protocol design and implementation. An area of such design concern includes routing. RPL has been proposed as a routing-protocol solution for WSN's by catering to the specific needs of low-power and lossy networks. Although RPL is emerging as a standard for routing in WSN's, it still faces many challenges concerning scalability with increases to network size, point-to-point traffic, and underspecification of handling node failures. In this thesis we propose Switching with Adaptive Interval Labels (SAIL), which takes a clean-slate approach from traditional routing. SAIL utilizes compact routing rather than destination-based routing, and caters to the issues RPL faces by replacing destination-based identifiers with interval labels. We implement SAIL using the ns-3 simulator and compare its performance to RPL in a wireless-network deployment. Our results show that SAIL outperforms RPL in energy conservation and forms shorter forwarding paths in small to medium-sized networks. For larger deployments we show that SAIL provides tremendously lower storage overhead where routing table sizes remain constant, while RPL routing tables grow linearly at O(n), where n equals the number of nodes in the network.

Efficient TCP enhancements for Hybrid Networks

By

Ramesh Srinivasan

The original TCP design and implementation still has widespread deployment today among the devices in the Internet, including various end-devices/clients ranging from main-frames, large servers, desktops, laptops, palm-tops, notebooks, notepads, smart-phones, tablets and now various IoT (Internet of Things) devices. The TCP architecture and design, most of it put in place in the 1980s and shortly thereafter prior to the Internet boom, was intended solely for a wired network. One of the main in-built assumptions, due to the above design, is that any packet drops or delays are interpreted as solely due to a congestion in the network and corresponding adjustments/changes are made by the protocol. However, with the advent of wireless networks and their co-existence with existing wired Internet deployment, the legacy TCP implementation ends up inadvertently responding to packet losses and increased latencies, due to wireless-link errors, as if it were due to network-congestion. This results in significantly reduced throughput and efficiencies.

WSC-TCP (Wireless Santa Cruz - TCP) proposed here, is an efficient TCP enhancement for today’s hybrid networks, which have both wireless as well as wired-links. It is a culmination of attempts to arrive at a framework to enable the TCP stack to differentiate those delays and losses, which are due to wireless-link-layer issues versus those that are due to actual network-congestion issues. The effort involved a study of the different mechanisms in the existing TCP design that do not work as per the original intent of responding only to network-congestion, due to its reliance on packet RTT (Round trip time) and related metrics for detecting and sizing the network congestion in today’s hybrid networks (wireless and wired links). The simulation tests conducted with some of the proposed enhancements result in an overall improved throughput/performance by 10-15% compared to TCP New-Reno in a hybrid network (wired and wireless links).

The network switch is one of the core components of computer communications, functioning as the interconnect among ever increasing number of nodes and a variety of networks. Whether it is used in data centers for connecting large numbers of nodes with high storage and computational capacities, or in the Internet backbone as a core router carrying vast quantity of information among its users, or in medium to large sized local area networks including educational campuses and companies, the switches require ever increasing design needs for scalability, high delivery ratios, low latencies, and quality-of-service (QoS) guarantees.

We believe that switch design should be evaluated as a holistic aspect of network protocol design, since many metrics in protocol design and switch design are closely correlated. Moreover, designing a switch requires the integration of hardware and software, because different functionalities of the switch can be implemented in hardware, software or both depending on different design considerations.

Rapid increase in Software Defined Networks (SDN) related research denotes that the approach of abstracting the data plane from the control plane, thus achieving higher flexibility and architectural simplicity, accounts for a notable research problem. Furthermore, label switching based routing algorithms enable the forwarding layer to be implemented at hardware and thus inflict less per-hop latency. Introducing a hardware component into network protocol design might initially seem to combine two orthogonal design components, however the outcome is an architecture that achieves multiple levels of abstraction, flexibility, simplicity, scalability and high performance at once.

In this thesis, we propose a novel architecture with an emphasis on hardware-software co-design paradigm, resulting in a scalable, flexible and high performance switch. We show that our design can be fully implemented on a Field-Programmable Gate Array (FPGA) based platform. Our system demonstrates an attempt to improve SDN, by taking the hardware component into account in the design flow, resulting in our Software-HArdware co-Defined nEtwork Switch (SHADES) for a label switched routing protocol, delivering load balancing, low latencies, and high delivery ratios.

Ad hoc wireless networks can be quickly deployed to provide wireless communication services in areas without pre-established infrastructure of base stations. Since delivering signals between sources and destinations is a primary function of ad hoc networks, the design and analysis of ad hoc networks are also a task of signal processing. In this paper, we review some of the major results on maximum throughput of ad hoc networks. Both single-hop ad hoc networks and multi-hop ad hoc networks are considered. In particular, we show examples of how to maximize the throughput of a large ad hoc network through a closed-loop analysis of two medium access control (MAC) schemes: synchronous array method (SAM) and asynchronous array method (AAM). Both omni-directional antennas and directional antennas are examined. This study illustrates an importance of the interplays between signal processing, information theory and networking protocols.