Dynamic Control of Data-Intensive Services over Edge Computing Networks
Yang Cai,Jaime Llorca,Antonia M. Tulino,Andreas F. Molisch
DOI: https://doi.org/10.48550/arXiv.2205.14735
2022-05-30
Abstract:Next-generation distributed computing networks (e.g., edge and fog computing) enable the efficient delivery of delay-sensitive, compute-intensive applications by facilitating access to computation resources in close proximity to end users. Many of these applications (e.g., augmented/virtual reality) are also data-intensive: in addition to user-specific (live) data streams, they require access to (static) digital objects (e.g., image database) to complete the required processing tasks. When required objects are not available at the servers hosting the associated service functions, they must be fetched from other edge locations, incurring additional communication cost and latency. In such settings, overall service delivery performance shall benefit from jointly optimized decisions around (i) routing paths and processing locations for live data streams, together with (ii) cache selection and distribution paths for associated digital objects. In this paper, we address the problem of dynamic control of data-intensive services over edge cloud networks. We characterize the network stability region and design the first throughput-optimal control policy that coordinates processing and routing decisions for both live and static data-streams. Numerical results demonstrate the superior performance (e.g., throughput, delay, and resource consumption) obtained via the novel multi-pipeline flow control mechanism of the proposed policy, compared with state-of-the-art algorithms that lack integrated stream processing and data distribution control.
Networking and Internet Architecture,Systems and Control