Scalable Identification of Load Imbalance in Parallel Executions Using Call Path Profiles

Tallent, Nathan R.; Adhianto, Laksono; Mellor-Crummey, John

doi:10.1109/sc.2010.47

Cited by 64 publications

(26 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Efficient, scalable measurement of load [12,24] identifies whether load imbalance is a problem for a particular application. Imbalance attribution [23] provides insight into the source code locations that cause imbalance. Our load model takes advantage of existing tools [10] and their measurements, and combines them with knowledge of the application elements and their interactions.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Quantifying the effectiveness of load balance algorithms

Pearce

Gamblin

Supinski

et al. 2012

Proceedings of the 26th ACM International Conference on Supercomputing

View full text Add to dashboard Cite

Load balance is critical for performance in large parallel applications. An imbalance on today's fastest supercomputers can force hundreds of thousands of cores to idle, and on future exascale machines this cost will increase by over a factor of a thousand. Improving load balance requires a detailed understanding of the amount of computational load per process and an application's simulated domain, but no existing metrics sufficiently account for both factors. Current load balance mechanisms are often integrated into applications and make implicit assumptions about the load. Some strategies place the burden of providing accurate load information, including the decision on when to balance, on the application. Existing application-independent mechanisms simply measure the application load without any knowledge of application elements, which limits them to identifying imbalance without correcting it.Our novel load model couples abstract application information with scalable measurements to derive accurate and actionable load metrics. Using these metrics, we develop a cost model for correcting load imbalance. Our model enables comparisons of the effectiveness of load balancing algorithms in any specific imbalance scenario. Our model correctly selects the algorithm that achieves the lowest runtime in up to 96% of the cases, and can achieve a 19% gain over selecting a single balancing algorithm for all cases.

show abstract

Section: Related Workmentioning

confidence: 99%

“…These tools can provide insight into the source location that caused an imbalance [23] and into the distribution of the load, but this knowledge alone is insufficient to correct the load. Existing load metrics do not account for constraints on rebalancing imposed by application elements and their interaction.…”

Section: Introductionmentioning

confidence: 99%

Quantifying the effectiveness of load balance algorithms

Pearce

Gamblin

Supinski

et al. 2012

Proceedings of the 26th ACM International Conference on Supercomputing

View full text Add to dashboard Cite

show abstract

“…CrayPat [3] calculates imbalance metrics from profiles that provide measures of absolute and relative load imbalance. HPC-TOOLKIT [23] attributes the costs of idleness at global synchronization points to overloaded call paths, highlighting imbalances in call-path profiles. However, profiling-based approaches, which aggregate performance data along the time dimension, can generally express only static imbalances reliably and do not capture dynamic load shifts between processes over time.…”

Section: Related Workmentioning

confidence: 99%

Scalable Critical-Path Based Performance Analysis

Böhme

Wolf

Supinski

et al. 2012

2012 IEEE 26th International Parallel and Distributed Processing Symposium

View full text Add to dashboard Cite

Abstract-The critical path, which describes the longest execution sequence without wait states in a parallel program, identifies the activities that determine the overall program runtime. Combining knowledge of the critical path with traditional parallel profiles, we have defined a set of compact performance indicators that help answer a variety of important performance-analysis questions, such as identifying load imbalance, quantifying the impact of imbalance on runtime, and characterizing resource consumption. By replaying event traces in parallel, we can calculate these performance indicators in a highly scalable way, making them a suitable analysis instrument for massively parallel programs with thousands of processes. Case studies with real-world parallel applications confirm that-in comparison to traditional profiles-our indicators provide enhanced insight into program behavior, especially when evaluating partitioning schemes of MPMD programs.

show abstract

“…Prior work has investigated the parallel measurement of per-process load balance data and its attribution to source code [6,10,18,19]. Two of these techniques complement this work by enabling scalable parallel data collection for the type of per-process data we collect here, and by allowing code to be automatically sliced into phases and regions based on callpaths.…”

Section: Related Workmentioning

confidence: 99%

Simplifying Performance Analysis of Large-scale Adaptive Scientific Applications

Bhatelé¹,

Gamblin²,

Gunney³

et al. 2012

View full text Add to dashboard Cite

Performance analysis of parallel scientific codes is becoming increasingly difficult due to the rapidly growing complexity of applications and architectures. Existing tools fall short in providing intuitive views that facilitate the process of performance debugging and tuning. In this paper, we exploit a recent idea of projecting and visualizing performance data on the communication and hardware domain for faster, more intuitive analysis of applications. We leverage several performance analysis and visualization tools to showcase the discovery of scalability bottlenecks in a structured AMR library. Using novel techniques to project per-phase timing data, application data, and communication data on a communication graph, we identify a previously elusive scaling bottleneck in the library. We present solutions that mitigate this problem, resulting in 22% improvement in the performance for a 65,536-core run on an IBM Blue Gene/P system.

show abstract

Scalable Identification of Load Imbalance in Parallel Executions Using Call Path Profiles

Cited by 64 publications

References 24 publications

Quantifying the effectiveness of load balance algorithms

Quantifying the effectiveness of load balance algorithms

Scalable Critical-Path Based Performance Analysis

Simplifying Performance Analysis of Large-scale Adaptive Scientific Applications

Contact Info

Product

Resources

About