A Vulnerability Factor for ECC-protected Memory

Abstract

Fault injection studies and vulnerability analyses have been used to estimate the reliability of data structures in memory. We survey these metrics and look at their adequacy to describe the data stored in ECC-protected memory. We also introduce FEA, a new metric improving on the memory derating factor by ignoring a class of false errors. We measure all metrics using simulations and compare them to the outcomes of injecting errors in real runs. This in-depth study reveals that FEA provides more accurate results than any state-of-the-art vulnerability metric. Furthermore, FEA gives an upper bound on the failure probability due to an error in memory, making this metric a tool of choice to quantify memory vulnerability. Finally, we show that ignoring these false errors reduces the failure rate on average by 12.75% and up to over 45%.

TD-NUCA: Runtime Driven Management of NUCA Caches in Task Dataflow Programming Models

“…The runtime system can transparently manage GPUs [7], [76], FPGA accelerators [18], [85], multi-node clusters [20], [27], [28], heterogeneous memories [4], [63], scratchpad memories [5], NUMA [81], [82] and cache coherent NUMA [21], [23] systems. Adding hardware support, the runtime system can guide cache replacement [37], [65], cache coherence deactivation [22], cache prefetching [47], [75], cache communication mechanisms in producer-consumer task relationships [64], [66], reliability and resilience [51]- [53], value approximation [19], and DVFS to accelerate critical tasks [26].…”

Runtime-Guided ECC Protection using Online Estimation of Memory Vulnerability

“…Furthermore, uncorrected errors in the same ECC word as accessed bits may cause crasheseven when the bits accessed are not erroneous themselves. Therefore, the most appropriate metric to quantify the vulnerability of an ECC word stored in memory is the vulnerability factor induced only by the memory accesses to this word [11,15]. While other proposals aim to build on or replace single-bit AVF with metrics based on memory access counts [10,36], all these existing approaches rely on offline profiling or simulating.…”

“…This paper focuses on the vulnerability in memory induced by memory accesses [11,15], to which we refer simply as vulnerability in the rest of this paper. For every ECC word in memory, cycles are categorised as either safe or vulnerable, depending on the next memory access to this word: cycles preceding a load are vulnerable, and cycles preceding a store are safe.…”

“…For this to be possible, we need to develop methodologies to automatically and dynamically quantify the vulnerability of the different portions of data stored in memory before a program executes. Prior work on measuring memory vulnerability relies on offline application profiling [10,11,15,36], but online methods are needed for widespread deployment.…”

Exploiting data locality in cache-coherent NUMA systems

“…Brumar et al [21] use the runtime information about the dependencies to build a memoisation approach to approximate the computations of executions and reduce the execution time. Jaulmes et al [72][73][74] evaluate the use of runtime systems in reliability and fault tolerance.…”