Systemwide Power Management with Argo

Ellsworth, Daniel; Patki, Tapasya; Perarnau, Swann; Seo, Sangmin; Amer, Amer; Zounmevo, Judicael A.; Gupta, Rinku; Yoshii, Kazutomo; Hoffman, Henry; Malony, Allen D.; Schulz, Martin; Beckman, Pete

doi:10.1109/ipdpsw.2016.81

Cited by 13 publications

(4 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We can outline the fact that a programming paradigm based on a graph of parallel and distributed tasks can obtain better performances on a huge amount of cores. Future supercomputers and post-petascale platforms would propose middle-ware and systems with smarter schedulers [12] and dedicated I/O systems [14] which will allow some efficient data persistence and anticipation of data migrations. In those machines, programming paradigms such as YML/XMP would be well-adapted and efficient.…”

Section: Discussionmentioning

confidence: 99%

“…Condor and PaRSEC are middlewares created for Grid then adapted to super-computers but they don't provide all the functionalities the next generation of supercomputer will need. Argo [12] is an operating system for extreme scale computing. It allows to reconfigure the resource nodes dynamically depending on a workload, to support massive concurrency, to present a hierarchical framework for power and fault management and a communication infrastructure.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Distributed and Parallel Programming Paradigms on the K computer and a Cluster

Gurhem

Tsuji

Petiton

et al. 2019

Proceedings of the International Conference on High Performance Computing in Asia-Pacific Region

View full text Add to dashboard Cite

In this paper, we focus on a distributed and parallel programming paradigm for massively multi-core supercomputers. We introduce YML, a development and execution environment for parallel and distributed applications based on a graph of task components scheduled at runtime and optimized for several middlewares. Then we show why YML may be well adapted to applications running on a lot of cores. The tasks are developed with the PGAS language XMP based on directives. We use YML/XMP to implement the block-wise Gaussian elimination to solve linear systems. We also implemented it with XMP and MPI without blocks. ScaLAPACK was also used to created an non-block implementation of the resolution of a dense linear system through LU factorization. Furthermore, we run it with different amount of blocks and number of processes per task. We find out that a good compromise between the number of blocks and the number of processes per task gives interesting results. YML/XMP obtains results faster than XMP on the K computer and close to XMP, MPI and ScaLAPACK on clusters of CPUs. We conclude that parallel and distributed multi-level programming paradigms like YML/XMP may be interesting solutions for extreme scale computing.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Distributed and Parallel Programming Paradigms on the K computer and a Cluster

Gurhem

Tsuji

Petiton

et al. 2019

Proceedings of the International Conference on High Performance Computing in Asia-Pacific Region

View full text Add to dashboard Cite

show abstract

“…Summarizing many of the available options, [31] discussed interactions between controls at different levels (server, rack, group of racks) with different objectives (energy efficiency, capping, performance) and actuators (p-states, turning machines off, admission control, power budget management). We also note [32] as an example of recent power management work for supercomputers.…”

Section: Related Workmentioning

confidence: 99%

Data Center Power Oversubscription with a Medium Voltage Power Plane and Priority-Aware Capping

Sakalkar

Kontorinis

Landhuis

et al. 2020

Proceedings of the Twenty-Fifth International Conference on Architectural Support for Programming Languages and Operating Syste

View full text Add to dashboard Cite

“…Again, this work can be done at the processor-level [23,24,44,56], DRAM [11], storage [30], and across a data-center [33,39]. At the data-center or cluster-level, power can be saved by consolidating workloads to use fewer physical machines [14,27,34,35,37,54], coordinating co-existing applications [45], and scheduling with green power [20] Scheduling jobs under a power cap has recently become a major concern for HPC operating systems [7,15] and job schedulers [3,19]. Recent work suggests that HPC workloads can actually achieve higher performance by over-provisioning large-scale installationssuch that using all nodes at full capacity would drastically violate the power budget-and severely power capping the individual nodes [47].…”

Section: Related Workmentioning

confidence: 99%

Performance & Energy Tradeoffs for Dependent Distributed Applications Under System-wide Power Caps

Zhang

Hoffmann

2018

Proceedings of the 47th International Conference on Parallel Processing

View full text Add to dashboard Cite

Large scale parallel machines are subject to system-wide power budgets (or caps). As these machines grow in capacity, they can concurrently execute dependent applications that were previously processed serially. Such application coupling saves IO and time as the applications now communicate at runtime instead of through disk. Such coupled applications are predicted to be a major workload for future exascale supercomputers; e.g., scientific simulations will execute concurrently with in situ analysis. While support for system-wide power caps has been widely studied, prior work does not consider the impact on coupled applications. We study techniques for maximizing coupled application performance under a system-wide power cap and implement them on a 26-node cluster. We compare to SLURM, a state-of-the-art job scheduler that considers power, but not coupling. The proposed techniques increase mean performance over SLURM by 7-14%. Unlike existing approaches, the proposed techniques also recognize when it is not possible to increase performance and, instead, reduce energy, achieving 18% energy reduction for a 5% performance loss. Finally, the dynamic techniques are resilient to tail behavior and system noise, improving performance in noisy environments by 30-36%.

show abstract

Systemwide Power Management with Argo

Cited by 13 publications

References 17 publications

Distributed and Parallel Programming Paradigms on the K computer and a Cluster

Distributed and Parallel Programming Paradigms on the K computer and a Cluster

Data Center Power Oversubscription with a Medium Voltage Power Plane and Priority-Aware Capping

Performance & Energy Tradeoffs for Dependent Distributed Applications Under System-wide Power Caps

Contact Info

Product

Resources

About