| There are many applications where a timely response to an important event is needed. Often such response can require significant computation and possibly communication, and it can be very challenging to complete it within the time-frame the response is needed. At the same time, there could be application-specific flexibility in the computation that may be desired. Also, the user could provide a benefit function, which captures what is most desirable to compute. Such applications are referred to as the adaptive application. Our goal of processing such applications is to optimize the benefit function (a Quality-of-Service metric) while satisfying certain constraints, such as time and resource budget.;In this thesis, we consider supporting such adaptive applications in grid and cloud environments . The applications could involve one or more adaptive parameters that can impact both application benefit and execution time. We first formulate the problem of parameter adaptation based on optimal control model and propose an autonomic adaptation algorithm. Then a middleware is presented to support such functionality and to enable development and deployment of the adaptive applications. Due to the resource heterogeneity in grid and cloud environments, performing such optimization further leads us to a resource selection and scheduling problem. We first consider the resource allocation problem in grids and define an efficiency value to reflect how effectively a particular service can be executed on a particular node, based on which we have developed a greedy scheduling algorithm to schedule these service components.;Another critical issue faced by the adaptive applications executed in the grid is the inherent unreliability of the resources. We next study fault tolerance for adaptive applications. Our approach is based on a multi-objective optimization algorithm for scheduling the application onto the most efficient and reliable resources. Furthermore, for the cases where failures do occur, we have developed a hybrid failure-recovery scheme to ensure that the application can complete within the pre-specified time interval.;The recent emergence of clouds is making the vision of utility computing realizable. Due to the pay-as-you-go model, the key consideration here is the trade-off between application benefit and resource budget. We present the design, implementation, and evaluation of a framework that can support such dynamic adaptation for applications in a cloud computing environment. The key component of our framework is a feedback control based dynamic resource provisioning algorithm. Furthermore, we also take the power management into account, when scheduling such scientific adaptive applications onto the cloud. We proposed pSciMapper, a power-aware consolidation framework for scientific workflows. The consolidation is viewed as a hierarchical clustering problem. We extract the key temporal features of the resource (CPU, memory, disk I/O, and network I/O) requirements of each workflow task, and use a dimensionality reduction method (KCCA) to relate the resource requirements to performance and power consumption.;We have extensively evaluated our proposed solutions using two real-world applications, i.e., Volume Rendering and Great Lake Nowcasting and Forecasting. The experimental results demonstrate the expected advantage of applying the proposed approaches. |
