Abstract

Distributed storage platforms support major systems that operate under high uncertainty but have strict, low-latency requirements. Such systems include health (remote surgery), finance (electronic trading), commerce (online shopping), cloud gaming, self-driving cars, and augmented reality. In most of these emerging applications, users’ instantaneous and expected numbers and their data interests extensively fluctuate. Therefore, traditional storage solutions, designed and provisioned to satisfy specific expected demands, are inadequate. This project first postulates a new metric that should be an essential consideration in designing efficient distributed storage that must provide low latency for time-sensitive applications and remain stable under high uncertainty. It then offers optimal storage solutions. Besides that, it advances fundamentals of coding theory, optimization on graphs, and real and finite geometry.

Redundancy increases the robustness of storage systems. Comparatively, erasure coding can efficiently service changing dynamics and requirements in data access. It is therefore essential to determine which data-access rates a redundant storage scheme can support. These rates make up its service-rate region. This project has two primary objectives: 1) For a given (implemented) redundancy scheme, it asks what its service-rate region is. 2) For a given (desired) service-rate region, it asks which redundancy scheme has the service-rate region that includes the desired one under some optimal cost or some required properties. In connection with the first question, the project is characterizing the service-rate regions of well-known classes of codes, as well as of those in use in storage systems to provide other desirable features. In connection with the second question, the project is designing codes to support desired service rates in resource-constrained settings.