Graph500: Traversing massive graphs with NAND Flash
Texas A&M University∗, Lawrence Livermore National Laboratory†
Roger Pearce∗, Maya Gokhale†, Nancy M. Amato∗
Abstract: We present our Graph 500 benchmark that uses NAND Flash to extend DRAM-based main memory for massive graph analysis. We apply our highly parallel asynchronous approach that hides data latency due to both poor locality and delays in the underlying graph data storage. We show our experiments from the June\’11 and November\’11 Graph500 list for both shared-memory multi-core systems with Fusion-io NAND Flash and distributed-memory clusters with node-local NAND Flash on trillion-edge datasets.
Enabling Application Directed Storage Devices
Princeton University∗, Fusion-io†
Anirudh Badam∗, David W. Nellans†
Abstract: A modern enterprise storage system is likely to contain: DRAM used as a page cache, a NAND-flash based block cache, and finally, a SAN composed of its own DRAM/Flash caching layers in front of an array of disk drives. Today, neither the OS nor the application has any control over how data is managed within these tiers. We propose a general purpose extension to the OS block layer that enables applications to express usage intent to generic block devices enabling performance optimization.
From UCSD 3rd Annual NVM Workshop: