Efficiently and quickly chewing through one trillion edges of a complex graph is no longer in itself a standalone achievement, but doing so on a single node, albeit with some acceleration and ultra-fast storage, is definitely worth noting.
A team from Georgia Tech has demonstrated Facebook and Microsoft-class capabilities for graph processing, but without the heavy-handed cluster approach via MOSAIC, a graph processing engine that exploits all the hardware resources available in a standard Xeon host processor, Xeon Phi coprocessors, NVMe, and a fast interconnect.
As one of the leads, Changwoo Min, tells The Next Platform, even though terascale graph processing is no longer unheard of, finding ways to make it efficient and high performance is where the real research and implementation challenge lies.
He points to the Microsoft work that spanned one trillion edges across a 64-node Infiniband connected cluster, but says with their single node on the same sized graph, the performance difference was only 1-2X slower on single-machine versus high-end cluster.
