A framework for providing hard delay guarantees and user fairness in Grid computing
Future Generation Computer Systems
| Hi-index | 0.00 |
The recent development of inexpensive high speed networking technology enables a computing environment where home applications and appliances can be connected to and run on shared general-purpose computing server(s), replacing traditional computing environments where each application or appliance has its own dedicated special purpose computing hardware. The advantages of this new computing environment over the traditional computing environment are lower cost, better resource utilization, and better manageability. However, there are many technical challenges that need to be solved in this new computing environment. This thesis is an effort to identify some of the technical challenges and to provide solutions. One key technical challenge is the level of computing resource guarantee that a shared general purpose server can provide to different devices, which have diverse computing resource requirements and varying performance requirements. This thesis presents the design and implementation of a reservation-based CPU management system, called dynamic soft real time (DSRT) system, on such a shared general purpose server. Even though it cannot provide a level of computing resource guarantee that is as good as dedicated special purpose hardware, the DSRT system can provide an acceptable level of guarantee to most soft real time (SRT) devices which can tolerate occasional violations of computing resource guarantees. The DSRT system distinguishes itself by providing new concepts: CPU service classes, multiprocessor partitions, and three-phase execution flow. It also provides probing service, adaptation service, advance reservation, distributed reservation, access control capability, and remote monitor to the client devices. In addition, the DSRT system has a platform-independent middleware design with multiprocessors support. This thesis further expands on the DSRT system to a multi-resource management system, called QoS-Aware Resource Management Framework (QualMan). QualMan manages various types of shared resources including the CPU, memory, and the network.