Access Type

Open Access Dissertation

Date of Award

January 2014

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Computer Science

First Advisor

Nathan W. Fisher

Abstract

Compositional real-time research has become one of the emerging trends in embedded and real-time systems due to the increasing scale and complexity of such systems. In this design paradigm, a large system is decomposed into smaller and simpler components, each of which abstracts their temporal requirements via interfaces. Such systems are mostly implemented by resource partitions to ensure that the components receive resources according to their interfaces. Potential implementations of a resource partition are via server-based interfaces or demand-based interfaces. In this context, our thesis in this dissertation is as follows:

Currently, server-based interfaces ensure strong temporal isolation among components at the cost of resource over-provisioning whereas demand-based interfaces precisely model the resource demand of a component without the guarantee of temporal isolation. For both these models, efficient and effective resource allocation as well as strict temporal isolation among components can be achieved. Specifically, we can obtain efficient and near-optimal bandwidth allocation schemes and admission controllers for periodic resource model and arbitrary demand-based interface respectively. Furthermore, efficient slack reclamation technique can be obtained to allocate unused processing resources at runtime while still enforcing the given interface.

To support our thesis, we address efficient resource allocation among components with server-based interfaces by providing fully-polynomial-time approximation schemes (FPTAS) for allocating processing resource to components scheduled by earliest-deadline-first (EDF) or fixed-priority (FP) scheduling algorithm. For enforcing temporal isolation of demand-based interfaces, we provide a parametric approximate admission control algorithm, which has polynomial-time complexity in terms of number of active jobs in the system and the approximation parameter. Finally, to address efficient reclamation of unused processing resources, we give a novel technique to optimally and efficiently determine maximum allowable runtime slack for a component with arbitrary interface, considering active jobs in the system and guaranteeing system schedulability even for worst-case future job arrival scenarios. We expect that these techniques can ultimately be used to minimize the size, weight, and power requirements of real-time and embedded systems by reducing the processing resource requirements of such systems.

Share

COinS