Title:
A sustainable software stack for parallel graph analysis

Abstract:
Graph theory is used to model large-scale complex systems in various domains, such as genomics and social network analysis. I will describe a sustainable software stack for parallel analysis of very large graphs. The layered software architecture combines performance, customizability, and conceptual simplicity though separation of concerns. I will first focus on the parallel backend, the Combinatorial BLAS, which implements state-of-the-art algorithms for sparse matrix primitives that serve as parallel building blocks for tightly-coupled graph computations. I will present the algorithmic innovations that make the primitives scalable, and the algebraic semiring model that is fundamental to the customizability of the Combinatorial BLAS. Several graph kernels that are implemented by composing a handful of the Combinatorial BLAS primitives scale up to tens of thousands of cores on distributed memory architectures, surpassing specialized implementations. I will then outline the architecture and capabilities of the Python based frontend, the Knowledge Discovery Toolbox, and describe several applications. I will briefly touch on supporting semantic graphs without sacrificing performance, thus enabling a broad set of real applications that maintain important auxiliary information through attributes on individual edges and vertices.

Bio:
Aydın Buluç is a Luis W. Alvarez Postdoctoral Fellow at Lawrence Berkeley National Laboratory and an affiliated researcher at UC Berkeley's Parallel Computing Laboratory. His research interests include parallel computing, combinatorial scientific computing, high performance graph analysis, and sparse matrix computations. He received his PhD in Computer Science from the University of California, Santa Barbara in 2010 and his BS in Computer Science and Engineering from Sabanci University, Turkey in 2005. As a graduate student, he spent a semester at the Mathematics Department of MIT, and a summer at the CSRI institute of Sandia National Labs in New Mexico.
He is a member of SIAM and ACM.