TAOBench: An End-to-End Benchmark for Social Networking Workloads

Download PDF

Abstract

The continued emergence of large social networking applications has introduced a scale of query volume and data that stresses the limits of existing data stores. However, few benchmarks accurately model these workload patterns, leaving researchers in short supply of tools to evaluate these systems and inform design choices. In this paper, we present a new benchmark, TAOBench, that captures the social graph workload at Meta (formerly known as Facebook). We open source workload configurations with a set of features that are sufficient for faithful reproduction of the social networking request patterns seen at Meta. We also build an open-source benchmark that can accurately model production workloads and generate new ones to test speculative scenarios. We validate our benchmark by running it at Meta and describe several uses cases. Furthermore, we report results for four popular distributed database systems. Our benchmark fills a gap in the available tools and data that researchers and developers can leverage to create the data stores of the future.