Design of a noninterfering debugger for embedded real-time systems.

Abstract

The problems of debugging parallel programs have been known for quite some time. However, the literature is suprisingly scarce in techniques for non-intrusive debugging of parallel programs. The issue of debugging parallel real-time programs introduces another dimension of difficulty, namely, scheduling which is an integral part of any real-time system. In a real-time system, it is possible that a program may miss its deadline because of an incorrect execution schedule between the processes of the program. The literature is also scarce in techniques for detecting and reproducing such errors. Various classes of errors that are typical of non real-time and real-time parallel programs are identified in this dissertation. For each class, the specific information that needs to be collected in order to reproduce it during the replay phase is identified. Also the sufficiency of information collected in order to reproduce the error during a subsequent execution is mathematically proved. A two-phase non-intrusive scheme for debugging parallel non real-time and real-time programs is described. The first phase involves monitoring the program and the second involves replaying the program in a simulated environment (while using the information collected during the first phase). Architectural support for collecting the above information in a non-intrusive manner is considered. The issue of non-intrusive checkpointing of the program to support a rollback and play back strategy of debugging is addressed. A novel memory architecture is proposed that facilitates such checkpointing while the program is being executed.