bootstrap responsive templates

ARM Research Summit BOF on
High Performance Graph Analytics: Algorithms, Programming, Architectures

Wednesday, September 18, 2019
11.15 AM - 12.45 PM
Sheraton Hotel at the Capitol, Austin, TX


Graph analytics represents a growing area of activity in government, industry and academia. This is clearly demonstrated by the increased number of DOE (E.g., ExaGraph and ExaLearn in the Exascale Computing Projects) and DOD projects (E.g., DARPA HIVE and SDH), as well as the appearance of a large variety of open source and commercial frameworks. Large-scale graph problems require ever growing compute power and impose significant requirements on modern high-performance computing architectures, from various points of view: memory, interconnect, processing elements, and their related specialization. Infrastructures like GraphLab, GraphX (on top of Spark), Pregel, Giraph, and, in general, vertex or edge-centric approaches based on MapReduce-like data-parallel models, have started to provide valuable results and found rapid adoption also on distributed memory systems, thanks to their higher-level programming abstractions, but have not demonstrated any significant ability to scale in size of the problems or performance on high-performance computing systems. 

This BOF aims at gathering the community of people interested in systems and frameworks for large-scale graph analytics, with the objective to discuss the current situation, identifying new challenges and opportunities, and laying the path towards future and interoperable infrastructures. 

Organizers: Antonino Tumeo (PNNL), Filippo Spiga (ARM)


Time Event
11.15 - 11.20Welcome and Introduction
Antonino Tumeo (PNNL), Filippo Spiga (ARM)
11.20 - 11.35Challenges and opportunities for graph algorithms and programming models
John Leidel (Tactical Computing Labs), Keshav Pingali (University of Texas, Austin)
11.35- 11.50Specializing Architectures for Data Analytics
David Donofrio (Tactical Computing Labs), Jason Riedy (Georgia Tech)
11.50 - 12.15Panel: "We can't build specialized architectures for graphs that can work efficiently with other workloads, so we just need to hand-optimize each and every algorithm for each and every architecture"
Moderator: Dough Joseph (ARM)
Panelists: David Donofrio (Tactical Computing Labs), Marco Minutoli (PNNL), Jason Riedy (Georgia Tech)
12.15 - 12.40Panel: "We can’t define appropriate abstractions for graph libraries, so MPI and OpenMP are our only solution"
Moderator: Antonino Tumeo (PNNL)
Panelists: John Leidel (Tactical Computing Labs), Keshav Pingali (University of Texas, Austin), Flavio Vella (Free University of Bozen)
12.40 - 12.45Wrap-up and Conclusions
Filippo Spiga (ARM), Antonino Tumeo (PNNL)