Bulk Synchronous Parallel Model on Graphic Processing Units

• Grant number: 12/23300-7
• Support Opportunities: Scholarships in Brazil - Doctorate
• Effective date (Start): March 01, 2013
• Effective date (End): February 28, 2018
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Alfredo Goldman vel Lejbman
• Grantee: Marcos Tulio Amaris González
• Host Institution: Instituto de Matemática e Estatística (IME). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Associated scholarship(s): 15/19399-6 - Machine learning to predict performance and running time of heterogeneous applications with uncertain data input, BE.EP.DR

Abstract

In this document a proposal in the specific topic of High Performance Computing in Computer Sciences is presented. GPUs devices were initially built for graphic computing, however, nowadays GPU devices are capable to perform more efficient parallel computation than multicore CPUs due their intrinsic parallel hardware architecture. Although five years ago, emergent application programming interfaces and programming languages introduced the concept General Purpose on GPU (GPGPU), researchers in this area have created applications with a level of parallelism in data and tasks that are able to run on mixed architectures of CPUs and GPUs. The Bulk Synchronous Parallel is a bridging model for parallel computation introduced by Valiant in 1990. The properties of the interconnection network are captured by a few architectural parameters for the BSP model in a parallel algorithm. Nevertheless, algorithms executed on massively parallel environments may face several inconveniences. Problems can arise, for instance, due to the balancing of the load or the communication latency. The main goal of this proposal is the development of models to perform efficiently BSP algorithms on parallel graphic processing unit architectures and it will be focused in the evaluation and characterization of the main problems that may negatively impact the implementation of efficient BSP algorithms on GPUs. The implemented algorithms based on the proposed models will be designed for massively parallel architectures, such as computer clusters, multicore systems, grid infrastructures, and in particular for GPU supercomputers.

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

BRUEL, PEDRO; AMARIS, MARCOS; GOLDMAN, ALFREDO. Autotuning CUDA compiler parameters for heterogeneous applications using the OpenTuner framework. CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE, v. 29, n. 22, SI, NOV 25 2017. (12/23300-7)

AMARIS, MARCOS; LUCARELLI, GIORGIO; MOMMESSIN, CLEMENT; TRYSTRAM, DENIS. Generic algorithms for scheduling applications on heterogeneous platforms. CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE, v. 31, n. 15, AUG 10 2019. (12/23300-7)

GENG, TONGSHENG; AMARIS, MARCOS; ZUCKERMAN, STEPHANE; GOLDMAN, ALFREDO; GAO, GUANG R.; GAUDIOT, JEAN-LUC. A Profile-Based AI-Assisted Dynamic Scheduling Approach for Heterogeneous Architectures. INTERNATIONAL JOURNAL OF PARALLEL PROGRAMMING, v. 50, n. 1, AUG 2021. (12/23300-7)

Academic Publications

(References retrieved automatically from State of São Paulo Research Institutions)

GONZÁLEZ, Marcos Tulio Amaris. Performance prediction of application executed on GPUs using a simple analytical model and machine learning techniques. 2018. Doctoral Thesis - Universidade de São Paulo (USP). Instituto de Matemática e Estatística (IME/SBI) São Paulo.