Increase in the usage of Internet of Things(IoT) devices across physical systems has provided a platform for continuous data collection, real-time monitoring, and extracting useful insights. Limited computing power and constrained resources on the IoT devices has driven the physical systems to rely on external resources such as cloud computing for handling compute-intensive and data-intensive processing. Recently, physical environments have began to explore the usage of edge devices for handling complex processing. However, these environments may face many challenges suchas uncertainty of device availability, uncertainty of data relevance, and large set of geographically dispersed devices. This research proposes the design of a reliable distributed management system that focuses on the following objectives: 1. improving the success rate of task completion in uncertain environments. 2. enhancing the reliability of the applications and 3. support latency sensitive applications. Main modules of the proposed system include: 1. A novel proactive user recruitment approach to improve the success rate of the task completion. 2.Contextual data acquisition and integration of false data detection for enhancing the reliability of the applications. 3. Novel distributed management of compute resources for achieving real-time monitoring and to support highly responsive applications. User recruitment approaches select the devices for offloading computation. Proposed proactive user recruitment module selects an optimized set of devices that match the resource requirements of the application. Contextual data acquisition module banks on the contextual requirements for identifying the data sources that are more useful to the application. Proposed reliable distributed management system can be used as a framework for offloading the latency sensitive applications across the volunteer computing edge devices.

Energy efficient design and management of data centers has seen considerable interest in the recent years owing to its potential to reduce the overall energy consumption and thereby the costs associated with it. Therefore, it is of utmost importance that new methods for improved physical design of data centers, resource management schemes for efficient workload distribution and sustainable operation for improving the energy efficiency, be developed and tested before implementation on an actual data center. The BlueTool project, provides such a state-of-the-art platform, both software and hardware, to design and analyze energy efficiency of data centers. The software platform, namely GDCSim uses cyber-physical approach to study the physical behavior of the data center in response to the management decisions by taking into account the heat recirculation patterns in the data center room. Such an approach yields best possible energy savings owing to the characterization of cyber-physical interactions and the ability of the resource management to take decisions based on physical behavior of data centers. The GDCSim mainly uses two Computational Fluid Dynamics (CFD) based cyber-physical models namely, Heat Recirculation Matrix (HRM) and Transient Heat Distribution Model (THDM) for thermal predictions based on different management schemes. They are generated using a model generator namely BlueSim. To ensure the accuracy of the thermal predictions using the GDCSim, the models, HRM and THDM and the model generator, BlueSim need to be validated experimentally. For this purpose, the hardware platform of the BlueTool project, namely the BlueCenter, a mini data center, can be used. As a part of this thesis, the HRM and THDM were generated using the BlueSim and experimentally validated using the BlueCenter. An average error of 4.08% was observed for BlueSim, 5.84% for HRM and 4.24% for THDM. Further, a high initial error was observed for transient thermal prediction, which is due to the inability of BlueSim to account for the heat retained by server components.