John Hosking

This study tackles the online user allocation problem in mobile edge computing (MEC) systems powered by non-orthogonal multiple access. App vendors need to determine a proper wireless channel in a base station/edge server and sufficient transmit power for every user. We consider a stochastic MEC system where users arrive and depart over time. When an edge server runs out of computing resources, some users will have to wait until the resources become available again, which incurs an allocation delay cost. This cost is often not investigated in many studies, which also do not consider a multi-cell, multi-channel system as we do in this work, due to its complexity. We aim to minimize the allocation delay and transmit power costs, increasing the system’s energy efficiency. To achieve this objective while guaranteeing users’ data rate requirements over time, we adopt the Lyapunov framework to convert this long-term optimization problem into a series of subproblems to be solved in every time slot. To solve the aforementioned subproblems efficiently, we present a distributed game theory-based approach. The proposed algorithm is theoretically evaluated and experimentally demonstrated to outperform several baseline and state-of-the-art methods, highlighting the significance of systematic consideration for both computation and communication aspects of this problem.

Mobile edge computing (MEC) allows edge servers to be placed at cellular base stations. App vendors like Uber and YouTube can rent computing resources and deploy latency-sensitive applications on edge servers for their users to access. Non-orthogonal multiple access (NOMA) is an emerging technique that facilitates the massive connectivity of 5G networks, further enhancing the capability of MEC. The edge user allocation (EUA) problem faces new challenges in 5G NOMA-based MEC systems. In this study, we investigate the EUA problem in a multi-cell multi-channel downlink power-domain NOMA-based MEC system. The main objective is to help mobile app vendors maximize their benefit by allocating maximum users to edge servers in a specific area at the lowest computing resource and transmit power costs. To this end, we introduce a decentralized game-theoretic approach to effectively select a channel and edge server for each user while fulfilling their resource and data rate requirements. We theoretically and experimentally evaluate our solution, which significantly outperforms various state-of-the-art and baseline approaches.

Mobile edge computing (MEC) is a new distributed computing paradigm where edge servers are deployed at, or near cellular base stations in close proximity to end-users. This offers computing resources at the edge of the network, facilitating a highly accessible platform for real-time, latency-sensitive services. A typical MEC environment is highly stochastic with random user arrivals and departures over time. Here, we address the user allocation problem from a service provider's perspective, who needs to allocate its users to the cloud or edge servers in a specific area. A user, who has a multi-dimensional resource requirement, can be allocated to either the remote cloud, which incurs a high latency, or an edge server, which results in a low latency but might require the user to wait in a queue. This article aims to achieve a controllable trade-off between performance (throughput) and several associated costs such as queuing delay and latency costs. We model this problem as a stochastic optimization problem, propose SUAC (Stochastic User AlloCation) – an online Lyapunov optimization-based algorithm, and prove its performance bounds. The experimental results demonstrate that SUAC outperforms existing approaches, effectively allocating users with a desired trade-off while keeping the system strongly stable.

Edge computing is a new distributed computing paradigm extending the cloud computing paradigm, offering much lower end-to-end latency, as real-time, latency-sensitive applications can now be deployed on edge servers that are much closer to end-users than distant cloud servers. In edge computing, edge user allocation (EUA) is a critical problem for any app vendors, who need to determine which edge servers will serve which users. This is to satisfy application-specific optimization objectives, e.g., maximizing users' overall quality of experience, minimizing system costs, and so on. In this paper, we focus on the cost-effectiveness of user allocation solutions with two optimization objectives. The primary one is to maximize the number of users allocated to edge servers. The secondary one is to minimize the number of required edge servers, which subsequently reduces the operating costs for app vendors. We first model this problem as a bin packing problem and introduce an approach for finding optimal solutions. However, finding optimal solutions to the NP-hard EUA problem in large-scale scenarios is intractable. Thus, we propose a heuristic to efficiently find sub-optimal solutions to large-scale EUA problems. Extensive experiments conducted on real-world data demonstrate that our heuristic can solve the EUA problem effectively and efficiently, outperforming the state-of-the-art and baseline approaches.

The majority of existing application profiling techniques aggregate and report performance costs by method or calling context. Modern large-scale object-oriented applications consist of thousands of methods with complex calling patterns. Consequently, when profiled, their performance costs tend to be thinly distributed across many thousands of locations with few easily identifiable optimisation opportunities. However experienced performance engineers know that there are repeated patterns of method calls in the execution of an application that are induced by the libraries, design patterns and coding idioms used in the software. Automatically identifying and aggregating costs over these patterns of method calls allows us to identify opportunities to improve performance based on optimising these patterns. We have developed an analysis technique that is able to identify the entry point methods, which we call subsuming methods, of such patterns. Our offline analysis runs over previously collected runtime performance data structured in a calling context tree, such as produced by a large number of existing commercial and open source profilers. We have evaluated our approach on the DaCapo benchmark suite, showing that our analysis significantly reduces the size and complexity of the runtime performance data set, facilitating its comprehension and interpretation. We also demonstrate, with a collection of case studies, that our analysis identifies new optimisation opportunities that can lead to significant performance improvements (from 20% to over 50% improvement in our case studies)

Real-time computation of data streams over affordable virtualized infrastructure resources is an important form of data in motion processing architecture. However, processing such data streams while ensuring strict guarantees on quality of services is problematic due to: (i) uncertain stream arrival pattern; (ii) need of processing different types of continuous queries; and (iii) variable resource consumption behavior of continuous queries. Recent work has explored the use of statistical techniques for resource estimation of SQL queries and OLTP workloads. All these techniques approximate resource usage for each query as a single point value. However, in data stream processing workloads in which data flows through the graph of operators endlessly and poses performance and resource demand fluctuations, the single point resource estimation is inadequate. Because it is neither expressive enough nor does it capture the multi-modal nature of the target data. To this end, we present a novel technique which uses mixture density networks, a combined structure of neural networks and mixture models, to estimate the whole spectrum of resource usage as probability density functions. The proposed approach is a flexible and convenient means of modeling unknown distribution models. We have validated the models using both the linear road benchmark and the TPC-H, observing high accuracy under a number of error metrics: mean-square error, continuous ranked probability score, and negative log predictive density.

Whilst robots are increasingly being deployed as social agents, it is still difficult to program them to interact socially. This is because current programming tools either require programmers to work at a low level or lack features needed to create certain aspects of social interaction. High level, domain specific tools with features designed specifically to meet the requirements of social interaction have the potential to ease the creation of social applications. We present a domain specific application programming interface (API) that is designed to meet the requirements of social interaction. The Cognitive Dimensions Framework was used as a design tool during the design process and the API was validated by implementing an exemplar application. The evaluation of the API showed that programmers with no robotics knowledge were positively impressed by the notation and that its organization, domain specific interfaces and object oriented nature positively affected several Cognitive Dimensions.