Efficient fault tolerance in multi-media applications through selective instruction replication

Abstract

As voltages decrease, soft errors are expected to become an increasing problem in maintaining program correctness. Unfortunately, previous mechanisms to improve processor reliability protect all processor instructions equally, causing such approaches to suffer from significant performance degradation and/or substantial hardware overhead. However, recent research has shown that in multimedia applications such as photography, video, and audio, not all instructions are created equal: many operations prove to be far more tolerant to faults than others [1].This observation can be leveraged to limit the cost of reliable computing by protecting only those instructions that are critical to correct execution. We propose a mechanism to protect against soft errors through selective instruction replication. We begin with a dynamic instruction replication framework that replicates every instruction and checks them upon commit, rolling back for any inconsistent results. Instead of replicating the entire program, instructions that the compiler identifies as tolerant to error would remain unprotected. While full replication requires 40% to 100% overhead, our mechanism requires only 30% to 75% overhead, reducing the overhead by 15-33% with minimal hardware overhead. We suffer only 0.5 -1% fidelity degradation with this approach.

Assuring application-level correctness against soft errors

“…Put together, our approach can lead to 22% average energy savings for multimedia applications while guaranteeing application-level correctness, when compared to a recent work [9], which cannot guarantee application-level correctness. Comparing to the technique proposed in [20] which guarantees both application-level and numerical correctness, our method achieves 79% energy reduction.…”

“…As can be seen, for the random error injections, the overhead of recovery is small primarily because of the relatively small number of critical instructions. Columns 6 and 7 show the energy savings achieved by our method compared to the approach in [9] -column 6 shows the impact of static analysis only while column 7 considers both static analysis and run-time monitoring. Column 8 shows the energy savings over the technique proposed in [20] that guarantees numerical correctness.…”

“…It should be noted the denominator used in computing the percentages is the corresponding value of the single error-free execution (without duplication) -hence, some of the percentages are abovc 100%. On the average, our technique provides 21% energy reduction compared to [9] and ensures application level correctness (which [9] may not guarantee) and 79% energy savings compared to [20] (which ensures numerical correctness and hence is a stricter form of checking). Columns 9 and 10 of Table 2 show the overhead associated with runtime monitoring (for the applications that benefited from runtime monitoring).…”

Selective SWIFT-R

“…They propose to transfer to the software world the concept of selective hardening, typical from the hardware world [22,39,40]. In [15,16], the authors propose the selective instruction replication to guarantee the application-level correctness in multimedia applications. This kind of applications can tolerate, in some cases, a execution which is not 100% numerically correct, and the program results can still appear to be correct from the user perspective [41].…”

“…In many cases, this is the main difficulty for the software-based techniques feasibility. In order to reduce these overheads and to offer more flexibility to designers, recent works have proposed the selective hardening based on software [14][15][16]. This consists of protecting only specific parts of the program or the microprocessor architectural resources (reachable from the instruction set architecture -ISA) by means of redundant software.…”