A data streaming model in MPI

Laure, Erwin; Holmes, Daniel J.; Bull, Mark

doi:10.1145/2831129.2831131

Cited by 18 publications

(6 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…A key example is StreamIt [124], a language that enables high performance of large streaming applications by efficiently mapping processes to a wide range of environments, including shared-memory architectures and HPC clusters. Another example is MPIStream, a prototype library implemented atop MPI, which provides an interface to existing MPI applications to adopt the streaming model [110,109]. Basically, MPIStream provides a lightweight approach to link MPI processes with different tasks by using four basic concepts of stream processing: communication channels, data producer/consumer, data streams and stream operations.…”

Section: Data Stream Processing Applications (Dasp)mentioning

confidence: 99%

Exploring stream parallel patterns in distributed MPI environments

et al. 2019

View full text Add to dashboard Cite

En primer lugar, a mis padres. Os debo todo lo que soy y seré. Gracias por el cuidado y educación recibidos, por los buenos (y malos) momentos, por aquellos días de piscina en Playa de Madrid, por las barbacoas en la Fundación del Lesionado Medular/ASPAYM Madrid, por las comidas en "La Bodeguita" y en el Centro Municipal de Mayores Ascao. Gracias a vosotros aprendí a ser fuerte. A mamá, porque me enseñaste lo más importante, a sonreir a pesar de la adversidad, y otras cosas de menor importancia, como mantener un hogar. A papá, por enseñarme gran parte de lo que sé, el equilibrio, e iluminarme el camino; aunque fuera de forma accidental, sin tí, mi vocación sería otra. Nunca os lo podré agradecer tanto como lo merecisteis. D.E.P.En segundo lugar, a mis compañeros de trabajo: Manu, Javi Doc, y David. Por todos los momentos compartidos en el laboratorio y fuera de él; por ayudarme y presionarme para que trabajara. Sin vosotros, dudo que estuviera escribiendo estas líneas ahora. A mis compañeros de laboratorio: Guille, Mario, Javi Prieto, y José Cabrero, por las charlas y momentos de risa. A Alejandro Calderón, que siempre ha estado ahí para escucharme y echar una mano. A José Daniel, por darme la oportunadad de trabajar en el grupo y por estar ahí en los momentos difíciles que últimamente han acontecido. También a mis compañeros de trabajo del CERN: Enric, Stephan, Guilherme, Bertrand, Javi, Emmanouil "Manos", y especialmente a Axel, Jakob y Pere, por acogerme en ROOT.No voy a terminar sin mencionar a las personas que siempre estuvieron ahí (no, no me olvido de vosotros): a mi hermano, a Úrsula, a mi tía Paloma y mi tío Matías, a mi primo "el niño".

show abstract

Section: Data Stream Processing Applications (Dasp)mentioning

confidence: 99%

Exploring stream parallel patterns in distributed MPI environments

et al. 2019

View full text Add to dashboard Cite

show abstract

“…The NCSw framework is divided in several abstract classes that represent the source of the input datasets and the target (or where) to conduct inference (Figure 3). The aim is to provide an easyto-use implementation that could enable the integration of new kinds of input sources (e.g., MPI streams [32]) or target devices (e.g., FPGA) in the future. The VPU implementation is based on the use of the NCAPI.…”

Section: Application Hostmentioning

confidence: 99%

Exploring the Vision Processing Unit as Co-Processor for Inference

Rivas-Gomez

Peña²,

Moloney

et al. 2018

2018 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)

Self Cite

View full text Add to dashboard Cite

The success of the exascale supercomputer is largely debated to remain dependent on novel breakthroughs in technology that effectively reduce the power consumption and thermal dissipation requirements. In this work, we consider the integration of co-processors in high-performance computing (HPC) to enable low-power, seamless computation offloading of certain operations. In particular, we explore the so-called Vision Processing Unit (VPU), a highly-parallel vector processor with a power envelope of less than 1W.We evaluate this chip during inference using a pre-trained GoogLeNet convolutional network model and a large image dataset from the ImageNet ILSVRC challenge. Preliminary results indicate that a multi-VPU configuration provides similar performance compared to reference CPU and GPU implementations, while reducing the thermal-design power (TDP) up to 8× in comparison.

show abstract

“…III. THE IMPLEMENTATION In this section, we introduce the implementation of the decoupling approach in an MPI-based stream library, called MP-IStream [12]. We choose this implementation approach because MPI is currently the de-facto programming system on supercomputers and is likely to be available on exascale systems [1].…”

Section: E Identify Operations For Decouplingmentioning

confidence: 99%

Preparing HPC Applications for the Exascale Era: A Decoupling Strategy

Gioiosa

Kestor

Laure

2017

2017 46th International Conference on Parallel Processing (ICPP)

Self Cite

View full text Add to dashboard Cite

Production-quality parallel applications are often a mixture of diverse operations, such as computation-and communication-intensive, regular and irregular, tightly coupled and loosely linked operations. In conventional construction of parallel applications, each process performs all the operations, which might result inefficient and seriously limit scalability, especially at large scale. We propose a decoupling strategy to improve the scalability of applications running on large-scale systems. Our strategy separates application operations onto groups of processes and enables a dataflow processing paradigm among the groups. This mechanism is effective in reducing the impact of load imbalance and increases the parallel efficiency by pipelining multiple operations. We provide a proof-of-concept implementation using MPI, the de-facto programming system on current supercomputers. We demonstrate the effectiveness of this strategy by decoupling the reduce, particle communication, halo exchange and I/O operations in a set of scientific and dataanalytics applications. A performance evaluation on 8,192 processes of a Cray XC40 supercomputer shows that the proposed approach can achieve up to 4× performance improvement.

show abstract

A data streaming model in MPI

Cited by 18 publications

References 20 publications

Exploring stream parallel patterns in distributed MPI environments

Exploring stream parallel patterns in distributed MPI environments

Exploring the Vision Processing Unit as Co-Processor for Inference

Preparing HPC Applications for the Exascale Era: A Decoupling Strategy

Contact Info

Product

Resources

About