Inference of Control Structures in Adaptive Networked Systems

A networked system is made up of three layers of abstraction: an infrastructure sub-system that realizes the physical & logical resources, a service-support system that maps the infrastructure components into usable objects, and an adaptation support system that decides on suitable allocation of resources to various applications to attain a desired QoS behavior. The system layers are under purview of different administrative regimes, interacting via inter-layer service interfaces (such as IaaS/SaaS layers in a cloud-hosted service). In this setting, the service-support system has multiple stake-holders: such as QoS adaptation agents, resource-aware end-users, QoS auditors, and reconfiguration managers. These stake-holders often need to dynamically infer the control structures employed in the service-support system (e.g., the topological placement of content caching sites in a CDN). Obtaining an accurate system tomography in the face of changing external environment conditions is critical to the realization of diverse stake-holder goals. The paper outlines the strategies and mechanisms to infer the system-internals for specific needs. Topology inferencing in a rate-adaptive video multicast network of content caching/rendering sites (e.g., NetFlix) is also described.