Unified Architectural Support for Soft-Error Protection or Software Bug Detection

Abstract

In this paper we propose a unified architectural support that can be used flexibly for either soft-error protection or software bug detection. Our approach is based on dynamically detecting and enforcing instructionlevel invariants. A hardware table is designed to keep track of run-time invariant information. During program execution, instructions access this table and compare their produced results against the stored invariants. Any violation of the predicted invariant suggests a potential abnormal behavior, which could be a result of a soft error or a latent software bug.In case of a soft error, monitoring invariant violations provides opportunistic soft-error protection to multiple structures in processor pipelines. Our experimental results show that invariant violations detect soft errors promptly and as a result, simple pipeline squashing is able to fix most of the detected soft errors. Meanwhile, the same approach can be easily adapted for software bug detection. The proposed architectural support eliminates the substantial performance overhead associated with software-based bug-detection approaches and enables continuous monitoring of production code.

GangES: Gang error simulation for hardware resiliency evaluation

“…For brevity, we present the cost vs. SDC reduction curves for a representative subset from our workload and metric combinations in Figure 9. 7 In the remainder of this section we use the term gap to signify the difference in the Y axis (SDC) for a given value on the X axis between the different curves.…”

“…Software anomaly based error detection has emerged as an attractive approach that detects only those hardware faults that propagate to software [7,10,13,16,21,24,27]. This approach places near-zero cost error monitors that watch for anomalous software behavior.…”

Trace-based microarchitecture-level diagnosis of permanent hardware faults

“…Racunas et al dynamically predict the valid set of values that an instruction will produce, and consider a departure from this prediction as a symptom of a (transient) fault [14]. Dimitrov and Zhou monitor the variance between the two most recent results produced by two dynamic instructions of the same PC, and any large variance indicates a possible soft error [5]. We have recently explored various symptoms -fatal traps, hangs, high OS activity, and program based invariant violations -as the mechanisms for detecting permanent and transient faults in our SWAT system [6], [15].…”

An empirical study of reported bugs in server software with implications for automated bug diagnosis

“…To this end, increasing the reliability of systems is becoming critically important. In spite of tremendous improvements in software engineering, testing and software reliability [34,29,30,22,31,24,33,16,27,14,13,12,7,26,36,38,40,21,15,11], many software bugs still escape testing and enter production. As others have noted [36], performing off-site analysis of production run failures at development sites has several limitations: 1) it is difficult to reproduce failures at the development site due to differences in the environment, 2) customers have privacy concerns over what information can be released for off-site diagnosis, and 3) the same bug may generate a different failure at multiple production sites; it is cumbersome for the development team to investigate every failure that occurs without any automated feedback regarding the root cause of the failure.…”