Programming Strategies for Irregular Algorithms on the Emu Chick

Abstract

The Emu Chick prototype implements migratory memory-side processing in a novel hardware system. Rather than transferring large amounts of data across the system interconnect, the Emu Chick moves lightweight thread contexts to near-memory cores before the beginning of each remote memory read. Previous work has characterized the performance of the Chick prototype in terms of memory bandwidth and programming differences from more typical, non-migratory platforms, but there has not yet been an analysis of algorithms on this system. This work evaluates irregular algorithms that could benefit from the lightweight, memory-side processing of the Chick and demonstrates techniques and optimization strategies for achieving performance in sparse matrix-vector multiply operation (SpMV), breadth-first search (BFS), and graph alignment across up to eight distributed nodes encompassing 64 nodelets in the Chick system. We also define and justify relative metrics to compare prototype FPGA-based hardware with established ASIC architectures. The Chick currently supports up to 68x scaling for graph alignment, 80 MTEPS for BFS on balanced graphs, and 50% of measured STREAM bandwidth for SpMV.

ButterFly BFS -- An Efficient Communication Pattern for Multi Node Traversals

“…Hein et al [28] review several different approaches to implement random memory accesses. Hein et al [29] and Belvirani et al [7] study different BFS algorithms on the Emu Chick. Hein et al [29] show that a BFS algorithm using remote writes is more efficient than using thread migrations.…”

“…Hein et al [29] and Belvirani et al [7] study different BFS algorithms on the Emu Chick. Hein et al [29] show that a BFS algorithm using remote writes is more efficient than using thread migrations. Belvirani et al [7] show that an Intel processor outperforms the Emu Chick while the Chick shows good scalability.…”

“…As graph sizes continue to grow and the need for close to real-time results are needed, optimizing the BFS traversal is especially important. Algorithms have been designed to exploit massively multi-threaded shared memory systems as the Cray XMT [2], NVIDIA GPU's [37], and the EMU Chick [29]. While many algorithms focus on the classical top-down BFS traversals, the introduction of the direction optimizing traversal first introduced by Beamer et al [4] has brought a plethora of BFS implementations.…”