Content area
With the growing amount of data generated by different data sources (e.g., sensor networks, text messages, mobile devices, etc.), processing this massive amount of data in various computing networks raises new problems. Providing resources of the computing network to multiple heterogeneous applications, like data processing applications with requested QoS, is a challenging issue in different computing network scenarios. The unique characteristics of different computing clusters make using a single resource provisioning algorithm not an efficient one-size-fits-all solution.
This dissertation focuses on the problem of resource provisioning for heterogeneous applications in different computing networks. It investigates how network-aware resource provisioning algorithms can help both network providers and tenants, in the two considered computing scenarios: (a) cloud computing networks, and (b) dispersed computing networks.
The contributions of this dissertation are as follows. First, we consider the hierarchical structure of the cloud computing network, which causes interdependent server failures. We propose ShadowBox and ECHO, two redundancy-aware cloud resource management systems. We show that they can significantly improve resource utilization, while keeping the availability of applications untouched. Next, we propose SPARCLE, a network-aware scheduler for stream processing applications in a dispersed computing network. SPARCLE shows that considering the unique characteristics of the dispersed computing network (e.g., heterogeneous computation and communication resources, limited connectivity, node/link failures, and resource fluctuations) can significantly increase the processing rate of stream processing applications.
