NCSI: Parallel Implementation of Block Matrix Operations in Distributed Computing Systems

Status	Completed
Mentor Name	Hao Ji
Mentor's XSEDE Affiliation	Education Allocation
Mentor Has Been in XSEDE Community	4-5 years
Project Title	Parallel Implementation of Block Matrix Operations in Distributed Computing Systems
Summary	This is a continuation of the previous XSEDE EMPOWER project "Novel Randomized Algorithms for Large-Scale Matrix Completion" conducted in summer 2018. In this project, we will develop, analyze, and implement block matrix computations using GraphX for distributed Spark systems, as well as their CUDA correspondents for distributed multiple-GPU systems. We will focus on a set of core matrix operations including sparse matrix multiplication, dense matrix multiplication, and Singular Value Decomposition, and evaluate their computational efficiency in many applications such as large-scale matrix completion and block Krylov subspace problems.
Job Description	The students will 1) evaluate the performance of the Spark implementation of matrix completion algorithm, compared to the alternating least squares (ALS) algorithm; 2) develop a new offloading strategy to support transferring large matrices from CPU to multiple GPUs; 3) implement block Conjugate Gradient algorithms using CUDA, based on the offloading strategy; and 4) evaluate the performance of block matrix operations, concerning weak scaling and strong scaling across distributed systems.
Computational Resources	XSEDE Bridges' GPUs, Hadoop and Spark resources
Contribution to Community
Position Type	Intern
Training Plan	I will have weekly meetings with the student to help him/her progressively build up the knowledge and skills in distributed multi-GPUs and Spark systems that are needed in this project. Please note that the student who worked on the summer project is already familiar with GraphX and Spark systems.
Student Prerequisites/Conditions/Qualifications	Must have an undergraduate at California State Polytechnic University, Pomona Must have a good understanding of linear algebra and good programming skills in C.
Duration	Semester
Start Date	08/23/2018
End Date	12/09/2018