A Performance and Scalability Analysis of the BlueGene/L Architecture

Davis, Kei; Hoisie, Adolfy; Johnson, Greg; Kerbyson, Darren J.; Lang, Mike; Pakin, Scott; Petrini, Fabrizio

doi:10.1109/sc.2004.8

Cited by 36 publications

(25 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Details on ASCI Q (Compaq) performance as well as how SAGE was actually used in the optimization of the system performance is described in [9]. The performance of Lightning (Linux) has been described in [10], and early details on the performance of Blue Gene/L (IBM) is given in [11].…”

Section: Parallel Scaling Performancementioning

confidence: 99%

The RAGE radiation-hydrodynamic code

et al. 2008

View full text Add to dashboard Cite

We describe RAGE, the "Radiation Adaptive Grid Eulerian" radiation-hydrodynamics code, including its data structures, its parallelization strategy and performance, its hydrodynamic algorithm(s), its (gray) radiation diffusion algorithm, and some of the considerable amount of verification and validation efforts. The hydrodynamics is a basic Godunov solver, to which we have made significant improvements to increase the advection algorithm's robustness and to converge stiffnesses in the equation of state. Similarly, the radiation transport is a basic gray diffusion, but our treatment of the radiation-material coupling, wherein we converge nonlinearities in a novel manner to allow larger timesteps and more robust behavior, can be applied to any multi-group transport algorithm.

show abstract

Section: Parallel Scaling Performancementioning

confidence: 99%

The RAGE radiation-hydrodynamic code

et al. 2008

View full text Add to dashboard Cite

show abstract

“…The techniques described by Petrini and provided by the HPCC benchmarking suite have been used to describe the network (and overall system) performance of many emerging supercomputer installations, including the Cray XT4 and Blue Gene/P at Oak Ridge National Laboratory [1], [2], the Blue Gene/L systems at Argonne and Lawrence Livermore National Laboratory [5], and the Roadrunner system at Los Alamos National Laboratory [3]. The HPCC benchmark is a prototypical example of a benchmarking suite that reports results as summary statistics.…”

Section: A Related Workmentioning

confidence: 99%

Diagnosing Anomalous Network Performance with Confidence

Settlemyer¹,

Hodson²,

Kuehn³

et al. 2011

View full text Add to dashboard Cite

Abstract-Variability in network performance is a major obstacle in effectively analyzing the throughput of modern high performance computer systems. High performance interconnection networks offer excellent best-case network latencies; however, highly parallel applications running on parallel machines typically require consistently high levels of performance to adequately leverage the massive amounts of available computing power. Performance analysts have usually quantified network performance using traditional summary statistics that assume the observational data is sampled from a normal distribution. In our examinations of network performance, we have found this method of analysis often provides too little data to understand anomalous network performance. Our tool, Confidence, instead uses an empirically derived probability distribution to characterize network performance. In this paper we describe several instances where the Confidence toolkit allowed us to understand and diagnose network performance anomalies that we could not adequately explore with the simple summary statistics provided by traditional measurement tools. In particular, we examine a multi-modal performance scenario encountered with an Infiniband interconnection network and we explore the performance repeatability on the custom Cray SeaStar2 interconnection network after a set of software and driver updates.

show abstract

“…While this restricts the programming and usage model of applications on BG/L, it has the benefit that applications can not be interrupted by background jobs or daemons. As a direct consequence, the machine is virtually noise free (Davis, Hoisie, Johnson, Kerbyson, Lang, Pakin, and Petrini 2004) This daemon manages the communication between the front end and compute nodes, executes system call requests from the compute nodes, and provides I/O access to applications. In addition, the CIOD allows tools to start and to control one additional I/O node daemon, which in turn can communicate with the CIOD and control the compute nodes using a proprietary debugging interface provided by the CIOD.…”

Section: Bg/l System Softwarementioning

confidence: 99%

BlueGene/L applications: Parallelism On a Massive Scale

Supinski

Schulz

Bulatov

et al. 2008

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

BlueGene/L (BG/L), developed through a partnership between IBM and Lawrence Livermore National Laboratory (LLNL), is currently the world's largest system both in terms of scale with 131,072 processors and absolute performance with a peak rate of 367 TFlop/s. BG/L has led the Top500 list the last four times with a Linpack rate of 280.6 TFlop/s for the full machine installed at LLNL and is expected to remain the fastest computer in the next few editions.However, the real value of a machine like BG/L derives from the scientific breakthroughs that real applications can produce by successfully using its unprecedented scale and computational power. In this paper, we describe our experiences with eight large scale applications on BG/L from several application domains, ranging from molecular dynamics to dislocation dynamics and turbulence simulations to searches in semantic graphs. We also discuss the challenges we faced when scaling these codes and present several successful optimization techniques. All applications show excellent scaling behavior, even at very large processor counts, with one code even achieving a sustained performance of more than 100 TFlop/s, clearly demonstrating the real success of the BG/L design.

show abstract

A Performance and Scalability Analysis of the BlueGene/L Architecture

Cited by 36 publications

References 2 publications

The RAGE radiation-hydrodynamic code

The RAGE radiation-hydrodynamic code

Diagnosing Anomalous Network Performance with Confidence

BlueGene/L applications: Parallelism On a Massive Scale

Contact Info

Product

Resources

About