On-Line Control Flow Error Detection and Correction Based on Monitoring Both Data-Flow and Control-Flow Graphs

This paper proposes two efficient online techniques, Control-flow and Data Errors Correction using Data-flow Graph Consideration and Miniaturized Check-Pointing, to correct control-flow errors. These techniques are based on: 1) addition of redundant codes in a given program; 2) monitoring both data-flow and control-flow graphs of the program to detect and correct control-flow and data errors. The major importance of these techniques is the ability to recover data errors, arisen due to occurrence of the control-flow errors, with imposing less performance and memory overheads compared to previous techniques. In order to evaluate the proposed techniques, a functional simulator is used, and behaviors of five well-known benchmarks are studied on a simple processor with a real operating system. The experimental results demonstrate that at least 93% and 90% of control-flow errors can be detected and corrected by CDCC and MCP, respectively. An optimization approach is also introduced to decrease memory and performance overheads.