Data decomposition of Monte Carlo particle transport simulations via tally servers

Romano, Paul K.; Siegel, A.; Forget, Benoit; Smith, Kord

doi:10.1016/j.jcp.2013.06.011

Cited by 20 publications

(24 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In the previous work by Romano et al, (7) estimates of the tally server parameters were made by analyzing a hypothetical depletion simulation of the Monte Carlo Performance Benchmark (8) on two target supercomputers: the Titan Cray XK7 at Oak Ridge National Laboratory and Intrepid Blue Gene/P at Argonne National Laboratory. For the sake of simplicity, some of the assumptions made in estimating these parameters were not conservative.…”

Section: Model Refinementsmentioning

confidence: 99%

“…Assuming the same physical quantities need to be tallied, d will not change. As in our previous work, (7) a range of d will be investigated.…”

Section: Target Modelmentioning

confidence: 99%

“…Table 1 gives a summary of the parameters used in the model predictions for the tally server overhead as well as those used in our previous work. (7) Number of annular regions …”

Section: Network Interconnectmentioning

confidence: 99%

“…Two decomposition methods have been proposed previously for addressing this problem: domain decomposition (4,5) and data decomposition. (6) In a recent paper, (7) Romano et al demonstrate an implementation of data decomposition via a tally server algorithm and show that it offers a viable means of performing full core light-water reactor simulations via Monte Carlo. A theoretical model was developed to predict the performance of a simulation using the tally server algorithm relative to a simulation based on full memory replication.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

On the use of tally servers in Monte Carlo simulations of light-water reactors

Romano

Forget

Smith

et al. 2014

SNA + MC 2013 - Joint International Conference on Supercomputing in Nuclear Applications + Monte Carlo

Self Cite

View full text Add to dashboard Cite

An algorithm for decomposing tally data in Monte Carlo simulations using servers has recently been proposed and analyzed. In the present work, we make a number of refinements to a theoretical performance model of the tally server algorithm to better predict the performance of a realistic reactor simulation using Monte Carlo. The impact of subdividing fuel into annular segments on parameters of the performance model is evaluated and shown to result in a predicted overhead of less than 20% for a PWR benchmark on the Mira Blue Gene/Q supercomputer. Additionally, a parameter space study is performed comparing tally server implementations using blocking and non-blocking communication. Non-blocking communication is shown to reduce the communication overhead relative to blocking communication, in some cases resulting in negative overhead.

show abstract

Section: Model Refinementsmentioning

confidence: 99%

“…Assuming the same physical quantities need to be tallied, d will not change. As in our previous work, (7) a range of d will be investigated.…”

Section: Target Modelmentioning

confidence: 99%

“…Table 1 gives a summary of the parameters used in the model predictions for the tally server overhead as well as those used in our previous work. (7) Number of annular regions …”

Section: Network Interconnectmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

On the use of tally servers in Monte Carlo simulations of light-water reactors

Romano

Forget

Smith

et al. 2014

SNA + MC 2013 - Joint International Conference on Supercomputing in Nuclear Applications + Monte Carlo

Self Cite

View full text Add to dashboard Cite

show abstract

“…For example, the method described in [16] stores tally data on a set of distributed "server" processes, to which tracking processes send tally writes via asynchronous MPI sends. Indeed, reasonable performance was observed in [17] for a variety of typical computer parameters with tallies at a scale relevant to full-core analyses.…”

Section: Introductionmentioning

confidence: 99%

Monte Carlo domain decomposition for robust nuclear reactor analysis

et al. 2014

Self Cite

View full text Add to dashboard Cite

Monte Carlo (MC) neutral particle transport codes are considered the gold-standard for nuclear simulations, but they cannot be robustly applied to high-fidelity nuclear reactor analysis without accommodating several terabytes of materials and tally data. While this is not a large amount of aggregate data for a typical high performance computer, MC methods are only embarrassingly parallel when the key data structures are replicated for each processing element, an approach which is likely infeasible on future machines. The present work explores the use of spatial domain decomposition to make full-scale nuclear reactor simulations tractable with Monte Carlo methods, presenting a simple implementation in a production-scale code. Good performance is achieved for mesh-tallies of up to 2.39TB distributed across 512 compute nodes while running a full-core reactor benchmark on the Mira Blue Gene/Q supercomputer at the Argonne National Laboratory. In addition, the effects of load imbalances are explored with an updated performance model that is empirically validated against observed timing results. Several load balancing techniques are also implemented to demonstrate that imbalances can be largely mitigated, including a new and efficient way to distribute extra compute resources across coarse domain meshes.

show abstract