Inverse Space-Filling Curve Partitioning of a Global Ocean Model

Dennis, John M.

doi:10.1109/ipdps.2007.370215

Cited by 39 publications

(29 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…POP currently supports three algorithms for distributing the blocks over the available MPI tasks, Cartesian, rake (Marquet and Dekeyser, 1998), and space-filling curve (Dennis, 2007). The Cartesian algorithm starts by organizing the tasks in a two-dimensional grid.…”

Section: Existing Block-partitioning Schemesmentioning

confidence: 99%

“…In addition, some tasks may be left with less work (or even no work) if one or more blocks assigned to it only contain land. As shown in Dennis (2007), load imbalance between tasks can significantly degrade the performance of high-resolution ocean simulations.…”

Section: Existing Block-partitioning Schemesmentioning

confidence: 99%

“…Typical performances are from one to a few model years per 24 h using thousands of cores (Dennis, 2007). Considering the fact that it takes at least 1000 yr to reach a nearstatistical-equilibrium state, innovations to increase the performance of these models and to efficiently analyse the data from the simulations have a high priority.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A distributed computing approach to improve the performance of the Parallel Ocean Program (v2.1)

Werkhoven¹,

Maassen

Kliphuis

et al. 2014

Geosci. Model Dev.

View full text Add to dashboard Cite

Abstract. The Parallel Ocean Program (POP) is used in many strongly eddying ocean circulation simulations. Ideally it would be desirable to be able to do thousand-yearlong simulations, but the current performance of POP prohibits these types of simulations. In this work, using a new distributed computing approach, two methods to improve the performance of POP are presented. The first is a blockpartitioning scheme for the optimization of the load balancing of POP such that it can be run efficiently in a multiplatform setting. The second is the implementation of part of the POP model code on graphics processing units (GPUs). We show that the combination of both innovations also leads to a substantial performance increase when running POP simultaneously over multiple computational platforms.

show abstract

Section: Existing Block-partitioning Schemesmentioning

confidence: 99%

Section: Existing Block-partitioning Schemesmentioning

confidence: 99%

See 1 more Smart Citation

A distributed computing approach to improve the performance of the Parallel Ocean Program (v2.1)

Werkhoven¹,

Maassen

Kliphuis

et al. 2014

Geosci. Model Dev.

View full text Add to dashboard Cite

show abstract

“…Some other algorithmic approaches also attempt to improve the parallel performance of ocean models. For example, a load-balancing algorithm based on the space-filling curve was proposed that not only eliminates land blocks but also reduces the communication overhead due to the reduced number of processes (Dennis, 2007;Dennis and Tufo, 2008). Beare and Stevens (1997) also proposed increasing the number of extra halos and communication overlaps in the parallel ocean general circulation.…”

Section: Introductionmentioning

confidence: 99%

P-CSI v1.0, an accelerated barotropic solver for the high-resolution ocean model component in the Community Earth System Model v2.0

Tang

Tseng

et al. 2016

Geosci. Model Dev.

View full text Add to dashboard Cite

Abstract. In the Community Earth System Model (CESM), the ocean model is computationally expensive for highresolution grids and is often the least scalable component for high-resolution production experiments. The major bottleneck is that the barotropic solver scales poorly at high core counts. We design a new barotropic solver to accelerate the high-resolution ocean simulation. The novel solver adopts a Chebyshev-type iterative method to reduce the global communication cost in conjunction with an effective block preconditioner to further reduce the iterations. The algorithm and its computational complexity are theoretically analyzed and compared with other existing methods. We confirm the significant reduction of the global communication time with a competitive convergence rate using a series of idealized tests. Numerical experiments using the CESM 0.1 • global ocean model show that the proposed approach results in a factor of 1.7 speed-up over the original method with no loss of accuracy, achieving 10.5 simulated years per wall-clock day on 16 875 cores.

show abstract

“…Therefore, efficient parallelization, which enables a model to utilize more processor cores for acceleration, becomes a technical focus in model development; a number of component models with efficient parallelization have sprung up. For example, the Community Ice CodE (CICE; Hunke and Lipscomb, 2008;Humpe et al, 2013) at 0.1 • horizontal resolution can scale to 30 000 processor cores on the IBM Blue Gene/L (Dennis and Tufo, 2008); the Parallel Ocean Program (POP; Kerbyson and Jones, 2005;Smith et al, 2010) at 0.1 • horizontal resolution can also scale to 30 000 processor cores on the IBM Blue Gene/L and 10 000 processor cores on a Cray XT3 (Dennis, 2007); the Community Atmosphere Model (CAM; Morrison and Gettelman, 2008;Neale et al, 2010Neale et al, , 2012) with a spectral element dynamical core (CAM-SE) at 0.25 • horizontal resolution can scale to 86 000 processor cores on a Cray XT5 (Dennis et al, 2012).…”

Section: Introductionmentioning

confidence: 99%

A new adaptive data transfer library for model coupling

et al. 2016

View full text Add to dashboard Cite

Abstract. Data transfer means transferring data fields from a sender to a receiver. It is a fundamental and frequently used operation of a coupler. Most versions of state-of-the-art couplers currently use an implementation based on the pointto-point (P2P) communication of the message passing interface (MPI) (referred to as "P2P implementation" hereafter). In this paper, we reveal the drawbacks of the P2P implementation when the parallel decompositions of the sender and the receiver are different, including low communication bandwidth due to small message size, variable and high number of MPI messages, as well as network contention. To overcome these drawbacks, we propose a butterfly implementation for data transfer. Although the butterfly implementation outperforms the P2P implementation in many cases, it degrades the performance when the sender and the receiver have similar parallel decompositions or when the number of processes used for running models is small. To ensure data transfer with optimal performance, we design and implement an adaptive data transfer library that combines the advantages of both butterfly implementation and P2P implementation.

show abstract

Inverse Space-Filling Curve Partitioning of a Global Ocean Model

Cited by 39 publications

References 18 publications

A distributed computing approach to improve the performance of the Parallel Ocean Program (v2.1)

A distributed computing approach to improve the performance of the Parallel Ocean Program (v2.1)

P-CSI v1.0, an accelerated barotropic solver for the high-resolution ocean model component in the Community Earth System Model v2.0

A new adaptive data transfer library for model coupling

Contact Info

Product

Resources

About