Minimizing Synchronization Overhead in the Implementation of MPI One-Sided Communication

Thakur, Rajeev; Gropp, William; Toonen, Brian

doi:10.1007/978-3-540-30218-6_15

Cited by 19 publications

(8 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…I.e., the amount of MPI RMA synchronization calls embedded is in direct proportion to that of MPI RMA communication calls. In fact, the MPI RMA synchronization operation adds a substantial overhead [11]. Hence, this method will definitely lead to great performance loss for applications with considerable RMA communication requests.…”

Section: Heat Conductionmentioning

confidence: 99%

Application Productivity and Performance Evaluation of Transparent Locality-aware One-sided Communication Primitives

Zhou

Gracia

2017

IJNC

View full text Add to dashboard Cite

Nowadays, the individual nodes of a distributed parallel computer consist of multi-or manycore processors allowing to execute more than one process per node. The large difference in communication speed within a node through shared memory, versus across nodes through the network interconnect, requires to use locality-aware communication schemes for any efficient distributed application. However, writing an efficient locality-aware MPI code is complex and error-prone, because the developer has to use very different APIs for communication operations within and across nodes, respectively, and manage inter-process synchronization. In this paper, we analyze and enhance a recent one-sided communication model, namely DART-MPI, which is implemented on top of MPI-3. In this runtime system, the complexities of handling locality-awareness of MPI memory access operations, either remote or local, and the related synchronization calls are hidden inside the related DART-MPI interfaces resulting in concise code and improved application and developer productivity. We have carried out in-depth evaluation of our DART-MPI system. Foremost, a micro benchmark is conducted to help understanding the prime performance overhead of implementing APIs in DART-MPI system, which is small and becomes negligible with the growing message sizes. We then compare the performance of DART-MPI and flat MPI without locality awareness, in particular blocking and non-blocking memory operations, using a realistic scientific application on a large-scale supercomputer. The comparison demonstrates that in most cases the DART-MPI version of this application shows better performance than the flat MPI version. Further, we compare the DART-MPI version to a functionally equivalent MPI version, which thus includes code to deal with data-locality, and show that DART-MPI realizes almost the full potential of highly optimized MPI while maintaining high productivity for non-expert programmers.

show abstract

Section: Heat Conductionmentioning

confidence: 99%

Application Productivity and Performance Evaluation of Transparent Locality-aware One-sided Communication Primitives

Zhou

Gracia

2017

IJNC

View full text Add to dashboard Cite

show abstract

“…Synchronous updates are extremely efficient in terms of execution speed especially as they can utilise two-sided communication using MPI to send and receive primitives. Asynchronous parallelisation schemes have to utilise one-sided communication primitives such as MPI "put" and "get", utilising the remote memory access mechanism, which, although slower, allows for the cell state to be asked for or provided on demand, without the need to wait on some eventual update [19].…”

Section: Equivalence Of Sequential and Parallel Implementationsmentioning

confidence: 99%

Probabilistic Pharmaceutical Modelling: A Comparison Between Synchronous and Asynchronous Cellular Automata

Bezbradica

Ruskin

Crane

2014

Parallel Processing and Applied Mathematics

View full text Add to dashboard Cite

Abstract. The field of pharmaceutical modelling has, in recent years, benefited from using probabilistic methods based on cellular automata, which seek to overcome some of the limitations of differential equation based models. By modelling discrete structural element interactions instead, these are able to provide data quality adequate for the early design phases in drug modelling. In relevant literature, both synchronous (CA) and asynchronous (ACA) types of automata have been used, without analysing their comparative impact on the model outputs. In this paper, we compare several variations of probabilistic CA and ACA update algorithms for building models of complex systems used in controlled drug delivery, analysing the advantages and disadvantages related to different modelling scenarios. Choosing the appropriate update mechanism, besides having an impact on the perceived realism of the simulation, also has practical benefits on the applicability of different model parallelisation algorithms and their performance when used in large-scale simulation contexts.

show abstract

“…Synchronous updates are extremely efficient in terms of execution speed especially as they can utilise two-sided communication using MPI "send" and "receive" primitives. Asynchronous parallelisation schemes have to use one-sided communication primitives such as MPI "put" and "get", utilising the remote memory access mechanism, which, although slower, allows for the cell state to be asked for or provided on demand, without the need to wait on some eventual update [16].…”

Section: Equivalence Of Sequential and Parallel Implementationsmentioning

confidence: 99%

Comparative analysis of asynchronous cellular automata in stochastic pharmaceutical modelling

Bezbradica

Ruskin

Crane

2014

Journal of Computational Science

View full text Add to dashboard Cite

In pharmaceutical modelling, cellular automata have been used as an established tool to represent molecular changes through discrete structural interactions. The data quality provided by such modelling is found suitable for the early drug design phase where flexibility is paramount. While both synchronous (CA) and asynchronous (ACA) types of automata have been used, analysis of their nature and comparative influence on model outputs is lacking. In this paper, we outline a representative probabilistic CA for modelling complex controlled drug formulations and investigate its transition from synchronous to asynchronous update algorithms. The key investigation points include quantification of model dynamics through three distinct scenarios, parallelisation performance and the ability to describe different release phenomena, namely erosion, diffusion and swelling. The choice of the appropriate update mechanism impacts the perceived realism of the simulation as well as the applicability of large-scale simulations.

show abstract

Minimizing Synchronization Overhead in the Implementation of MPI One-Sided Communication

Cited by 19 publications

References 9 publications

Application Productivity and Performance Evaluation of Transparent Locality-aware One-sided Communication Primitives

Application Productivity and Performance Evaluation of Transparent Locality-aware One-sided Communication Primitives

Probabilistic Pharmaceutical Modelling: A Comparison Between Synchronous and Asynchronous Cellular Automata

Comparative analysis of asynchronous cellular automata in stochastic pharmaceutical modelling

Contact Info

Product

Resources

About