Abstract

Current wireless mobile networks are not able to support next generation applications that require low latency or produce large volumes of data that can overwhelm the network. Examples include video analysis applications, wearable devices, safety critical applications and intelligent smart city systems. The use of servers on the wide-area cloud, however, is also not an option as these applications require low response times, or involve processing of large volumes of data from many devices. To address these challenges, edge computing proposes the addition of computation and storage capabilities to the edge of the network [30, 19]. This thesis generalizes edge computing into a hierarchical cloud architecture deployed over the geographic span of a network. The vision supports scalable processing by providing storage and computation along a succession of datacenters positioned between the end device and the traditional wide area cloud datacenter. I develop a new deployment and execution platform called CloudPath based on the Function as a Service (FaaS) model that supports code and data mobility and distribution by enforcing a clear separation between computation and state. In CloudPath applications will be composed of a collection of light-weight stateless event handlers that can be implemented using high level languages, such as Java. In this thesis, I also develop a shared database abstraction called PathStore that enables transparent data access to the hierarchy of cloud and edge datacenters. PathStore supports concurrent object reads and writes on all nodes of the database hierarchy and its extension called SessionStore adds session consistency (read your own writes, monotonic reads/writes) for mobile applications. Finally, I implement a geo-distributed query engine that exploits the hierarchical structure of our eventually-consistent geo-distributed database to trade temporal accuracy (freshness) for improved latency and reduced bandwidth.

 

Manuscript

 

Bibtex