Measuring and Comparing the Scaling Behaviour of a High-Performance CFD Code on Different Supercomputing Infrastructures

Frisch, Jérôme; Mundani, Ralf-Peter

doi:10.1109/synasc.2015.63

Cited by 2 publications

(1 citation statement)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our fluid flow simulation framework is custom-tailored to be deployed on current top tier massive parallel machines. It has been tested on multiple high-performance clusters and showed good scaling behaviour on up to 140, 000 cores [2]. Nevertheless, we also identified bottlenecks, one of which turned out to be the central management instance.…”

Section: Foundationsmentioning

confidence: 94%

Distributed Domain Generation for Large-Scale Scientific Computing

Ertl

Mundani

2021

2021 20th International Symposium on Parallel and Distributed Computing (ISPDC)

Self Cite

View full text Add to dashboard Cite

In this work, we present methods for distributed domain generation within the constraints of our decentral domain management concept. Here, all participating actors only have knowledge of their immediate neighbours, which are defined by geometric and hierarchical relations between nodes that represent subsets of the computational domain. We generate this domain following a hierarchical spacetree refinement. First, an initial tree is generated on every participating process. Second, this tree is distributed following a space-filling curve linearisation locally. Every process is assigned at least one leaf node of the initial tree, which acts as a starting point for the subsequent domain generation. From here, every process independently refines a subdomain using a decomposition method, which transforms a triangular surface-based geometry description into a volumebased one, using increasingly complex intersection tests. The resulting domain tree is distributed, yet neighbourhood references of neighbouring subtrees are not resolved. We combine the resolution of these relations with a 2:1 tree balancing, which involves the transfer of the surface of neighbouring subtrees. We provide results of a domain generation testcase, using an input geometry with 84,072 triangles on up to 896 processes of the CoolMUC-2 cluster segment of LRZ's Linux Cluster System. Here, we bring down the overall time it takes to generate an adaptively refined and balanced octree with depth d = 7 from 5.5 hours on one process to two seconds on 896 processes.

show abstract

Section: Foundationsmentioning

confidence: 94%

Distributed Domain Generation for Large-Scale Scientific Computing

Ertl

Mundani

2021

2021 20th International Symposium on Parallel and Distributed Computing (ISPDC)

Self Cite

View full text Add to dashboard Cite

show abstract

Design and optimisation of an efficient HDF5 I/O Kernel for massive parallel fluid flow simulations

Ertl

Frisch

Mundani

2017

Concurrency and Computation

Self Cite

View full text Add to dashboard Cite

Summary More and more massive parallel codes running on several hundreds of thousands of cores are entering the computational science and engineering domain, allowing high‐fidelity computations on up to trillions of unknowns for very detailed analyses of the underlying problems. Such runs typically produce gigabytes of data, hindering both efficient storage and (interactive) data exploration. Advanced approaches based on inherently distributed data formats such as hierarchical data format version 5 become necessary here to avoid long latencies when storing the data and to support fast (random) access when retrieving the data for visual processing. This paper shows considerations and implementation aspects of an I/O kernel based on hierarchical data format version 5 that supports fast checkpointing, restarting, and selective visualisation using a single shared output file for an existing computational fluid dynamics framework. This functionality is achieved by including the framework's hierarchical data structure in the file, which also opens the door for additional steering functionality. Finally, the performance of the kernel's write routines are presented. Bandwidths close to the theoretical peak on modern supercomputing clusters were achieved by avoiding file‐locking and using collective buffering.

show abstract

Measuring and Comparing the Scaling Behaviour of a High-Performance CFD Code on Different Supercomputing Infrastructures

Cited by 2 publications

References 22 publications

Distributed Domain Generation for Large-Scale Scientific Computing

Distributed Domain Generation for Large-Scale Scientific Computing

Design and optimisation of an efficient HDF5 I/O Kernel for massive parallel fluid flow simulations

Contact Info

Product

Resources

About