Publications

Results 76–100 of 206

Search results

Jump to search filters

Asking the right questions: Benchmarking fault-tolerant extreme-scale systems

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Widener, Patrick W.; Ferreira, Kurt; Levy, Scott; Bridges, Patrick G.; Arnold, Dorian; Brightwell, Ronald B.

Much recent research has explored fault-tolerance mechanisms intended for current and future extreme-scale systems. Evaluations of the suitability of checkpoint-based solutions have typically been carried out using relatively uncomplicated computational kernels designed to measure floating point performance. More recent investigations have added scaled-down "proxy" applications to more closely match the composition and behavior of deployed ones. However, the information obtained from these studies (whether floating point performance or application runtime) is not necessarily of the most value in evaluating resilience strategies. We observe that even when using a more sophisticated metric, the information available from evaluating uncoordinated checkpointing using both microbenchmarks and proxy applications does not agree. This implies that not only might researchers be asking the wrong questions, but that the answers to the right ones might be unexpected and potentially misleading. We seek to open a discussion on whether benchmarks designed to provide predictable performance evaluations of HPC hardware and toolchains are providing the right feedback for the evaluation of fault-tolerance in these applications, and more generally on how benchmarking of resilience mechanisms ought to be approached in the exascale design space. © 2014 Springer-Verlag Berlin Heidelberg.

More Details
TYPE Conference YEAR 2014
ScopusOSTI

The portals 4.0.1 network programming interface

Barrett, Brian B.; Brightwell, Ronald B.; Pedretti, Kevin; Hemmert, Karl S.

This report presents a specification for the Portals 4.0 network programming interface. Portals 4.0 is intended to allow scalable, high-performance network communication between nodes of a parallel computing system. Portals 4.0 is well suited to massively parallel processing and embedded systems. Portals 4.0 represents an adaption of the data movement layer developed for massively parallel processing platforms, such as the 4500-node Intel TeraFLOPS machine. Sandias Cplant cluster project motivated the development of Version 3.0, which was later extended to Version 3.3 as part of the Cray Red Storm machine and XT line. Version 4.0 is targeted to the next generation of machines employing advanced network interface architectures that support enhanced offload capabilities. 3

More Details
TYPE SAND Report YEAR 2013
DOIOSTI
Results 76–100 of 206
Results 76–100 of 206