An end-to-end qos management architecture for flexible applications

Several applications in real-time systems are changing from being rigid monoliths to being composed of smaller interdependent components. These application components, each with tunable output performance (or quality), are connected end-to-end to provide service to end users. We observe that often the individual components are flexible they can trade off among the amounts of time and resources they use, their available input quality, and the output quality they produce. The end-to-end application itself is then flexible. Support infrastructure for such applications must (a) take advantage of the execution flexibility that a component may offer while (b) guaranteeing end-to-end quality of service (QoS) requirements. This thesis describes a distributed quality of service (QoS) management architecture for such end-to-end flexible applications. We describe a middleware based on this architecture that supports the specification and maintenance of application QoS while meeting end-to-end QoS (including temporal) guarantees. Through QoS negatiation, the middleware determines the quality levels and resource allocations of application components. Our performance studies focus on the design tradeoffs of negotiation policies. We also derive an optimal resource allocation strategy to meet end-to-end latency guarantees for CPU-bandwidth based resource allocation.