Advanced Algorithms in Heterogeneous and Uncertain Networking Environments

TL;DR

This paper addresses the challenge of optimizing network performance in heterogeneous and uncertain environments by analyzing sources of uncertainty, reviewing existing solutions, proposing new algorithms with performance guarantees, and validating them through implementation and testing.

Contribution

It introduces general design principles and practical algorithms for managing heterogeneity and uncertainty in networks, supported by rigorous analysis and empirical evaluation.

Findings

Algorithms achieve worst-case performance guarantees.

Practical algorithms outperform existing solutions in tests.

Insights into handling uncertainty across network layers.

Abstract

Communication networks are used today everywhere and on every scale: starting from small Internet of Things (IoT) networks at home, via campus and enterprise networks, and up to tier-one networks of Internet providers. Accordingly, network devices should support a plethora of tasks with highly heterogeneous characteristics in terms of processing time, bandwidth energy consumption, deadlines and so on. Evaluating these characteristics and the amount of currently available resources for handling them requires analyzing all the arriving inputs, gathering information from numerous remote devices, and integrating all this information. Performing all these tasks in real time is very challenging in today's networking environments, which are characterized by tight bounds on the latency, and always-increasing data rates. Hence, network algorithms should typically make decisions under uncertainty. This work addresses optimizing performance in heterogeneous and uncertain networking environments. We begin by detailing the sources of heterogeneity and uncertainty and show that uncertainty appears in all layers of network design, including the time required to perform a task; the amount of available resources; and the expected gain from successfully completing a task. Next, we survey current solutions and show their limitations. Based on these insights we develop general design concepts to tackle heterogeneity and uncertainty, and then use these concepts to design practical algorithms. For each of our algorithms, we provide rigorous mathematical analysis, thus showing worst-case performance guarantees. Finally, we implement and run the suggested algorithms on various input traces, thus obtaining further insights as to our algorithmic design principles.