DAGuE: A Generic Distributed DAG Engine for High Performance Computing

Bosilca, George; Bouteiller, Aurélien; Danalis, Anthony; Hérault, Thomas; Lemarinier, Pierre; Dongarra, Jack

doi:10.1109/ipdps.2011.281

Cited by 157 publications

(180 citation statements)

References 20 publications

(10 reference statements)

Supporting

Mentioning

174

Contrasting

Unclassified

Order By: Relevance

“…In terms of speedup values the above improvements imply an additional speedup improvement of 1.28 and 1.23 for our two different test cards respectively. These achievements are comparable to the ones presented in other recent works in the literature [14,[16][17]. Furthermore, the value θ that leads to the larger improvement for 4 CPU-cores is 0.8 for the GTX 760 and 0.9 for the GTX TitanX GPU, whereas the corresponding values for 8 CPU-cores become 0.6 and 0.8 respectively.…”

Section: ) Performance Of the Hybrid Openmp+cuda Schemesupporting

confidence: 87%

“…It also includes an automated distribution of the computational load on the CPU and the GPU, and achieves very good performance mainly in audio-processing systems. In [14,[16][17] more recent, relevant approaches are presented in the field of linear algebra and systems. In [17] an additional speedup of up to 1.25 is achieved (compared to the GPU-only implementation) for the parallel execution of the conjugate gradient method, whereas in [14,16] the CPU/GPU collaboration schemes in the field of linear algebra achieved a speedup ranging from 1.15 up to 1.35 for different sizes and types of problems.…”

Section: Related Workmentioning

confidence: 99%

“…In [14,[16][17] more recent, relevant approaches are presented in the field of linear algebra and systems. In [17] an additional speedup of up to 1.25 is achieved (compared to the GPU-only implementation) for the parallel execution of the conjugate gradient method, whereas in [14,16] the CPU/GPU collaboration schemes in the field of linear algebra achieved a speedup ranging from 1.15 up to 1.35 for different sizes and types of problems. The works of [14,17] have the major advantage of supporting dynamic distribution of the workload between CPU and GPU.…”

Section: Related Workmentioning

confidence: 99%

“…In [17] an additional speedup of up to 1.25 is achieved (compared to the GPU-only implementation) for the parallel execution of the conjugate gradient method, whereas in [14,16] the CPU/GPU collaboration schemes in the field of linear algebra achieved a speedup ranging from 1.15 up to 1.35 for different sizes and types of problems. The works of [14,17] have the major advantage of supporting dynamic distribution of the workload between CPU and GPU. Finally in [39] the authors present a novel generic framework that transparently orchestrates collaborative execution of a single data-parallel kernel across multiple asymmetric CPUs and GPUs.…”

Section: Related Workmentioning

confidence: 99%

“…Although in simplex parallelization there have not been noticed as many relevant attempts as one would expect and no parallel GPU-based implementation of the simplex algorithm has yet offered significantly better performance relative to an efficient sequential simplex solver (at least not in all types of problems), quite significant progress has been achieved at least for dense LP problems [12]. There also exist various approaches in the field of parallel scientific computing that adopt extended CPU/GPU collaboration [13][14][15][16][17]. However, the efficient CPU/GPU collaboration through the combined use of relevant programming models (such as OpenMP and CUDA) still remains a major research challenge.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Simplex Parallelization in a Fully Hybrid Hardware Platform

Mamalis¹,

Perlitis²

2017

ijacsa

View full text Add to dashboard Cite

Abstract-The simplex method has been successfully used in solving linear programming (LP) problems for many years. Parallel approaches have also extensively been studied due to the intensive computations required, especially for the solution of large LP problems. Furthermore, the rapid proliferation of multicore CPU architectures as well as the computational power provided by the massive parallelism of modern GPUs have turned CPU / GPU collaboration models increasingly into focus over the last years for better performance. In this paper, a highly scalable implementation framework of the standard full tableau simplex method is first presented, over a hybrid parallel platform which consists of multiple multicore nodes interconnected via a high-speed communication network. The proposed approach is based on the combined use of MPI and OpenMP, adopting a suitable column-based distribution scheme for the simplex tableau. The parallelization framework is then extended in such a way that it can exploit concurrently the full power of the provided resources on a multicore single-node environment with a CUDA-enabled GPU (i.e. using the CPU cores and the GPU concurrently), based on a suitable hybrid multithreading/GPU offloading scheme with OpenMP and CUDA. The corresponding experimental results show that the hybrid MPI+OpenMP based parallelization scheme leads to particularly high speed-up and efficiency values, considerably better than in other competitive approaches, and scaling well even for very large / huge linear problems. Furthermore, the performance of the hybrid multithreading/GPU offloading scheme is clearly superior to both the OpenMP-only and the GPU-only based implementations in almost all cases, which validates the worth of using both resources concurrently. The most important, when it is used in combination with MPI in a multi-node (fully hybrid) environment, it leads to substantial improvements in the speedup achieved for large and very large LP problems.

show abstract

Section: ) Performance Of the Hybrid Openmp+cuda Schemesupporting

confidence: 87%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Simplex Parallelization in a Fully Hybrid Hardware Platform

Mamalis¹,

Perlitis²

2017

ijacsa

View full text Add to dashboard Cite

show abstract

Data‐driven execution of fast multipole methods

Ltaief

Yokota

2013

Concurrency and Computation

View full text Add to dashboard Cite

SUMMARYFast multipole methods (FMMs) have scriptO (N) complexity, are compute bound, and require very little synchronization, which makes them a favorable algorithm on next‐generation supercomputers. Their most common application is to accelerate N‐body problems, but they can also be used to solve boundary integral equations. When the particle distribution is irregular and the tree structure is adaptive, load balancing becomes a non‐trivial question. A common strategy for load balancing FMMs is to use the work load from the previous step as weights to statically repartition the next step. The authors discuss in the paper another approach based on data‐driven execution to efficiently tackle this challenging load balancing problem. The core idea consists of breaking the most time‐consuming stages of the FMMs into smaller tasks. The algorithm can then be represented as a directed acyclic graph where nodes represent tasks and edges represent dependencies among them. The execution of the algorithm is performed by asynchronously scheduling the tasks using the queueing and runtime for kernels runtime environment, in a way such that data dependencies are not violated for numerical correctness purposes. This asynchronous scheduling results in an out‐of‐order execution. The performance results of the data‐driven FMM execution outperform the previous strategy and show linear speedup on a quad‐socket quad‐core Intel Xeon system.Copyright © 2013 John Wiley & Sons, Ltd.

show abstract

Faithful performance prediction of a dynamic task‐based runtime system for heterogeneous multi‐core architectures

Stanisic

Thibault

Legrand

et al. 2015

Concurrency and Computation

View full text Add to dashboard Cite

SUMMARYMulti-core architectures comprising several graphics processing units (GPUs) have become mainstream in the field of high-performance computing. However, obtaining the maximum performance of such heterogeneous machines is challenging as it requires to carefully off-load computations and manage data movements between the different processing units. The most promising and successful approaches so far build on task-based runtimes that abstract the machine and rely on opportunistic scheduling algorithms. As a consequence, the problem gets shifted to choosing the task granularity, task graph structure, and optimizing the scheduling strategies. Trying different combinations of these different alternatives is also itself a challenge. Indeed, obtaining accurate measurements requires reserving the target system for the whole duration of experiments. Furthermore, observations are limited to the few available systems at hand and may be difficult to generalize. In this article, we show how we crafted a coarse-grain hybrid simulation/emulation of StarPU, a dynamic runtime for hybrid architectures, over SimGrid, a versatile simulator of distributed systems. This approach allows to obtain performance predictions of classical dense linear algebra kernels accurate within a few percents and in a matter of seconds, which allows both runtime and application designers to quickly decide which optimization to enable or whether it is worth investing in higher-end graphics processing units or not. Additionally, it allows to conduct robust and extensive scheduling studies in a controlled environment whose characteristics are very close to real platforms while having reproducible behavior.

show abstract

DAGuE: A Generic Distributed DAG Engine for High Performance Computing

Cited by 157 publications

References 20 publications

Simplex Parallelization in a Fully Hybrid Hardware Platform

Simplex Parallelization in a Fully Hybrid Hardware Platform

Data‐driven execution of fast multipole methods

Faithful performance prediction of a dynamic task‐based runtime system for heterogeneous multi‐core architectures

Contact Info

Product

Resources

About