Real-Time Adaptive Resource Management

D istributed mission-critical environments employ a mixture of hard and soft real-time applications that usually expect a guaranteed range of quality of service. These applications have different levels of criticality and varied structures ranging from periodic independent tasks to distributed pipelines or event-driven modules. The underlying distributed system must evolve and adapt to the high variability in resource demands that competing applications impose. The current industry trend is to use commercial off-the-shelf (COTS) hardware and software components to build distributed environments for mission-critical applications. Adding a middleware layer above the COTS components facilitates consistent management of system resources, decreases system complexity, and reduces development costs. The real-time adaptive resource management system consists of a middleware layer that provides integrated services for real-time mission-critical distributed applications. RTARM includes a number of features that facilitate distributed resource management: • scalable end-to-end criticality-based QoS contract negotiation, which allows distributed applications to share common resources while maximizing their use and execution quality; • end-to-end QoS adaptation that dynamically adjusts resource use according to availability; • integrated services for CPU, network , and other resources with end-to-end QoS guarantees; • real-time application QoS monitoring for integrated services, and • plug-and-play components for easy extensibility for new services. As Figure 1 shows, RTARM's hierarchical architecture includes service managers , recursive structural entities that encapsulate a set of resources and their management mechanism. A higher-level service manager may receive a request for an integrated service that requires resources from lower-level service managers. Lower-level service managers directly control resources such as the CPU, network, or memory. The entire hierarchy treats resources and negotiation requests uniformly. This hierarchy allows dynamic configuration, since new service managers can join the system at any time. The admission protocol builds a virtual spanning tree over the service manager hierarchy that remains valid throughout the application's lifetime. The service manager hierarchy forms a directed acyclic graph, with service managers as nodes and edges represented by the " uses-services-from " relation. The virtual spanning tree built over the service manager hierarchy allows the service managers to conduct QoS translation and reverse translation. A hierarchical, recursive resource management architecture facilitates implementing QoS representations on top of basic services in complex distributed applications. A richer QoS representation simplifies application design and facilitates consistent resource management for incompatible applications. Regardless of the complexity of the application architecture and the QoS semantics at the top of the service manager …