Age of information for real-time network applications

Driven by recent advances in ubiquitous connectivity and pervasive computing, real-time status updates to interested recipients have become increasingly popular in streaming applications. These status updating systems all share a common need: the recipients want their information about the sources to be as fresh as possible. This thesis aims to analyze a recently proposed information freshness/timeliness metric, age of information (AoI), in various real-time network applications, and optimize the corresponding AoI metric given the network constraints. In this thesis, we model the real-time status updating system as source-receiver pairs connected through the networks. The first fundamental problem we consider is how timely update messages should be compressed based on the given network capacity. Different from traditional data compression techniques that shorten the average codeword length, we show that the optimal lossless compression scheme for fast message delivery depends on higher moments of the codeword length due to the queueing delay. The AoI-optimal codebook can be constructed by a recursive search algorithm based on the convex AoI penalty function. In ultra-dense network deployments, real-time updates are expected to be distributed to massive numbers of receivers via the nearby storage nodes at the network edge. Thus, the second fundamental problem we address is how the real-time updates should be replicated and distributed to multiple edge storage nodes through multicast networks. We answer the question by evaluating the average AoI at a receiver which has access to a random number of edge storage nodes, given the distribution of the random network delay. This system model is also applicable to time-sensitive content updates in Dynamo-type distributed storage systems in which the write and read operations go to multiple storage nodes simultaneously. We derive the AoI-optimal quorum mechanism that balances the data consistency and operation latency. Beyond the study of the two fundamental problems in AoI, we extend the similar AoI analysis to applications with resource contention and scheduling. We examine a remote cache updating system in which the local server maintains snapshots of the content at different remote sources and updates those snapshots according to a constrained rate. We compare AoI to an alternative Age of Synchronization freshness metric and evaluate the optimal rate allocation scheme for the two different age metrics. We also examine the edge cloud computational offloading system with multiple users, and investigate the scheduling policy for incoming jobs in a vision processing application at an edge server. We show that a greedy scheduling policy is optimal for a class of AoI-related penalty functions.Ph.D.Includes bibliographical reference