Engineering Papers⌕ Search

DOE OSTI · 964332

A Programming Model for Massive Data Parallelism with Data Dependencies

Abstract

Accelerating processors can often be more cost and energy effective for a wide range of data-parallel computing problems than general-purpose processors. For graphics processor units (GPUs), this is particularly the case when program development is aided by environments such as NVIDIA s Compute Unified Device Architecture (CUDA), which dramatically reduces the gap between domain-specific architectures and general purpose programming. Nonetheless, general-purpose GPU (GPGPU) programming remains subject to several restrictions. Most significantly, the separation of host (CPU) and accelerator (GPU) address spaces requires explicit management of GPU memory resources, especially for massive data parallelism that well exceeds the memory capacity of GPUs. One solution to this problem is to transfer data between the GPU and host memories frequently. In this work, we investigate another approach. We run massively data-parallel applications on GPU clusters. We further propose a programming model for massive data parallelism with data dependencies for this scenario. Experience from micro benchmarks and real-world applications shows that our model provides not only ease of programming but also significant performance gains.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Cui, Xiaohui, Mueller, Frank, Potok, Thomas E, Zhang, Yongpeng. 2009-01-01. A Programming Model for Massive Data Parallelism with Data Dependencies. https://www.osti.gov/biblio/964332

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related reports

Design and Integration of High Precision Superconducting Magnet Power Supply Systems

This paper reviews the design and integration approach being taken to power more than 400 superconducting magnets in Electron Ion Collider (EIC) by power supplies ranging from 20V to 400V and 100A to 18kA. A major challenge is to integrate existing legacy power supplies with new high current systems and maximize performance and reduce costs. Successful implementation requires coordinated integration of power convertors, current regulation, quench protection, energy extraction, machine protection, controls and existing accelerator infrastructure.

43 PARTICLE ACCELERATORS↗

Searching for the Most Harmful Field Errors in the HSR IR Superconducting Magnets

In this project, we improve beam stability for the Electron-Ion Collider. Magnetic field errors can reduce beam stability, making it essential to identify the field errors that have the greatest impact on accelerator performance. However, this is particularly challenging because beam stability depends on the complex interactions of many magnetic field errors, resulting in a high-dimensional and nonlinear optimization problem. We determine which field errors are the most influential for the large physical aperture superconducting magnet B2PF, a critical magnet in the Interaction Region (IR) in the Hadron Storage Ring (HSR). We complete and analyze nearly 30,000 simulations on the Brookhaven National Laboratory Linux Cluster by varying 18 nonlinear magnetic field errors. We evaluate beam stability using the dynamic aperture and the tune diffusion. We identify the field errors that most strongly influence beam stability and establish quantitative field error tolerances that improve accelerator performance.

43 PARTICLE ACCELERATORS↗