Analytics on Graphs with a Trillion Edges

Big graphs occur naturally in many applications, most obviously in social networks, but also in many other areas such as biology and forensics. Current approaches to processing large graphs use either supercomputers or very large clusters. In both cases the entire graph must reside in memory before it can be processed. We are pursuing an alternative approach, processing graphs from secondary storage. While this comes with some performance penalty, it makes analytics on very large graphs feasible on a small number of commodity machines. It also has the pleasing property that "if you can store a graph, you can compute on it". 

We have developed two systems, one for a single machine and one for a cluster of machines. X-Stream, the single-machine solution, aims to make all secondary storage access sequential. It uses two techniques to achieve this goal: edge-centric processing and streaming partitions. X-Stream outperforms the state-of-the-art GraphChi system, because it achieves better sequentiality and because it requires less preprocessing. Slipstream, the cluster solution, starts from the observation that there is little benefit to locality when accessing secondary storage over a high-speed network. As a result, we use lightweight dynamic partitioning, focusing on achieving load balance and sequential access to secondary storage. The resulting system achieves good scaling and outperforms other systems. With Slipstream we have also been able to process a trillion-edge graph, a new milestone for graph size on a small cluster. I will describe both systems and their performance on a number of benchmarks and in comparison to the state-of-the-art alternatives.

This work is joint work with Laurent Bindschaedler, Jasmina Malicevic and Amitabha Roy.