Abstract
In order to achieve the best application‐level Quality‐of‐Service (QoS), multimedia applications need to be dynamically tuned and reconfigured to adapt to fluctuating computing and communication environments. QoS‐sensitive adaptations are critical when applications run in general‐purpose systems, with no mechanisms provided for supporting resource reservations and real‐time guarantees. Such adaptations are triggered by resource availability variations caused by best‐effort resource allocations in unpredictable open environments.
In this paper, we argue that adaptations are most effective to achieve a better QoS when performed within applications, where they may be optimized towards the best performance tradeoffs across various application parameters with different semantics. However, we believe that decisions about when and how adaptations should occur need to be coordinated, and formalized as a generic algorithm to be applied to a wide range of applications. For this purpose, we first identify an application model to focus on a set of application‐specific tuning ‘knobs’ and critical parameters, then propose a polynomial‐complexity QoS probing algorithm to quantitatively capture the run‐time relationships between the two sets of parameters. Finally, we present a hierarchical adaptive QoS control architecture to bridge the gap between original ‘triggers’ of adaptation and actual tuning ‘knobs’ to be invoked. To prove the validity of our architecture and algorithms, we present Agilos, a middleware implementation of our hierarchical architecture. Under its control, we show that a configurable multimedia tracking application is able to deliver optimal performance even when operating in unpredictable open environments.