Design and Optimization of Heterogeneous Networks
thesisposted on 01.08.2019, 00:00 by Shanyu Zhou
The motto of Internetting everything, everywhere, all the time is becoming reality thanks to the increasing number and diversity of devices with Internet connectivity such as smartphones, tablets, wearable devices, and connected vehicles. However, existing resource allocation and optimization mechanisms do not address the full range of challenges, specifically heterogeneity due to the diverse and dynamic nature of these devices. The goal of this thesis is to develop new networking mechanisms that make better use of available resources by taking into account the heterogeneity. We first focus on transportation systems of heterogeneously connected vehicles, where not all vehicles are able to communicate due to equipment limitation, privacy and security concerns, and lossy communication channels. We develop (i) a connectivity-aware traffic phase scheduling algorithm that increases the average number of vehicles that are allowed to pass the intersection, and (ii) a shortest routing algorithm with minimum traveling delay between two points in a transportation network. Then we consider a wireless network where an arbitrary number of heterogeneous queues (per-flow and FIFO queues) with arbitrary number of flows share the same wireless medium. We develop a resource allocation scheme; dFC, and a stochastic flow control and scheduling algorithm; qFC, that significantly improve the throughput of the wireless network, as compared to the well-known queue-based flow control and max-weight scheduling. Furthermore, we consider a cellular network with heterogeneous priority traffic (foreground vs background) for Internet of Things. We develop in-network traffic controller for low priority background traffic that quickly yields to foreground traffic when the network is congested, and quickly recaptures spare capacity when it becomes lightly loaded. Lastly, we focus on distributed computing systems with hard deadlines where a computation intensive task is divided into sub-tasks and allocated to heterogeneous helper devices. We explore the benefit of (n,k) MDS coded computation over uncoded computation and develop optimal k that maximizes the probability of meeting the deadline.