Big Data Workloads Drawn from Real-Time Analytics Scenarios Across Three Deployed Solutions

摘要

Big Data solution vendors and customers alike face a pressing need for a few credible benchmarking workloads for demonstrating or optimizing performance, elasticity, efficiency, and robustness of solutions they create or deploy. Many new problems require extraction of immediately actionable intelligence from torrents of data, so a good application level benchmark must reflect in its design both real-time (low latency) and high throughput metrics. It should also impose loads that reflect the realities of complex, interdependent mixes of storage and analysis operations. This short paper describes three different application level scenarios. In these scenarios Big Data solutions are used to generate answers in real time for a subset of requests while requests that do not require such real time responses are completed at high rate in the background in presence of massive inflows of new data. The solutions from which we draw these scenarios are already in deployment or in pre-deployment testing, and thus can serve as good models from which to draw design perspectives in assembling a realistic Big Data workload, meaningful to customers tackling realtime needs while balancing high availability and service rate requirements.