A Large-Scale Study of Failures in High-Performance Computing Systems

By Bianca Schroeder, Garth A. Gibson. IEEE Transactions on Dependable and Secure Computing, 2006

1,086 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Information Systems

View Paper: https://www.semanticscholar.org/paper/168b8cbbbacd234f23b70b952ef58b5b56e67529

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Information Systems

Citations in the corpus

ProBO: Versatile Bayesian Optimization Using Any Probabilistic Programming Language (2019, background): “Examples of this are queries involving unstable simulations [29], or faulty computer systems [41].”
Evaluating the viability of process replication reliability for exascale systems (2011, methodology): “Based on past study results [37], system MTTI decreases linearly with increased system socket count.”
Addressing failures in exascale computing (2014, methodology): “Results of studies show variance, however; a study by Schroeder and Gibson in 2010 showed that failures attributable to hardware ranged from 30%–60% (Schroeder and Gibson, 2010).”
DRAM errors in the wild: a large-scale field study (2009, background): “In production sites running large-scale systems, memory component replacements rank near the top of component replacements [20] and memory errors are one of the most common hardware problems to lead to machine crashes [19].”
Memory Errors in Modern Systems: The Good, The Bad, and The Ugly (2015, background): “Accelerated testing on the SRAM used in Hopper was performed using a variety of particle sources, including the high-energy neutron beam at the LANSCE ICE House[22] at LANL and an alpha particle source.”
Design, Modeling, and Evaluation of a Scalable Multi-level Checkpointing System (2010, background): “…Moody, Greg Bronevetsky, Kathryn Mohror and Bronis R. de Supinski Lawrence Livermore National Laboratory Livermore, CA 94551 USA {moody20, bronevetsky, kathryn, bronis}@llnl.gov Abstract High-performance computing (HPC) systems are growing more powerful by utilizing more hardware components.”
Availability Knob: Flexible User-Defined Availability in the Cloud (2016, background): “Many studies [41, 57, 58] have shown that the time between failure in large computing systems is well modeled by the Weibull distribution.”
WorkflowSim: A toolkit for simulating scientific workflows in distributed environments (2012, background): “What’s more, many researchers [5][6][7][8][9][10] have emphasized the importance of fault tolerant design and concluded that the failure rates in modern distributed systems should not be neglected.”
FTI: High performance Fault Tolerance Interface for hybrid systems (2011, background): “In the exceptional case of losing more data than tolerated by the L2 scheme, FTI will then search for the last checkpoint saved in the PFS. Finally, FTI_Finalize() check that all the FT-managers have ﬁn-ished their job and free all the resources.”
An Analysis of Traces from a Production MapReduce Cluster (2010, background): “Other studies of production traces of computer systems and clusters have also focused on failure data and characteristics [25]–[29].”
A Survey on Automated Log Analysis for Reliability Engineering (2020, background): “lity issues. 3 LOGGING Logging is the act to collect numerical and textual data of software behavior, such as low-memory conditions or attempts to access a file. Logging is valuable for both academia [110, 119, 126, 144] and industry [9, 18, 50, 119, 152, 156] across a variety of application domains”
CPR: Understanding and Improving Failure Tolerant Training for Deep Learning Recommendation with Partial Recovery (2020, methodology): “…training jobs in production datacenter ﬂeets with failures, our study ﬁnds that the mean-time-between-failures (MTBF) was 14–30 hours on average, similar to the statistics from other production-scale systems (Schroeder & Gibson, 2009; Kondo et al., 2010; Garraghan et al., 2014; Sahoo et al., 2004).”