Haohan Li

Embedded Software Engineer, Ph.D.

Meta Reality Labs
1 Hacker Way,
Menlo Park, CA 94025

Email: lihaohan#cs.unc.edu

Title

Abstract

This dissertation addresses the following question to the design of scheduling policies and resource allocation mechanisms in contemporary embedded systems that are implemented on integrated computing platforms: in a multitasking system where it is hard to estimate a task's worst-case execution time, how do we assign task priorities so that 1) the safety-critical tasks are asserted to be completed within a specified length of time, and 2) the non-critical tasks are also guaranteed to be completed within a predictable length of time if no task is actually consuming time at the worst case?

This dissertation tries to answer this question based on the mixed-criticality real-time system model, which defines multiple worst-case execution scenarios, and demands a scheduling policy to provide provable timing guarantees to each level of critical tasks with respect to each type of scenario. Two scheduling algorithms are proposed to serve this model. The OCBP algorithm is aimed at discrete one-shot tasks with an arbitrary number of criticality levels. The EDF-VD algorithm is aimed at recurrent tasks with two criticality levels (safety-critical and non-critical). Both algorithms are proved to optimally minimize the percentage of computational resource waste within two criticality levels. More in-depth investigations to the relationship among the computational resource requirement of different criticality levels are also provided for both algorithms.

Final Documents

Dissertation [pdf]
Defense presentation slides [pdf, ppsx]

Back to main page