How to Maximize Database Concurrency

High concurrency is crucial for optimal database performance, but overwhelming the database with excessive requests can lead to decreased throughput and increased latency. In a recent webinar, Brian Taylor from Optimizely discussed how to maximize database concurrency effectively by understanding the Universal Scaling Law (USL), which describes how system throughput changes with varying user numbers. The USL identifies three regions: linear, saturation, and retrograde, with the linear region being where throughput is directly proportional to concurrency, while the saturation region sees constant throughput, and the retrograde region experiences decreasing throughput with increased concurrency. To address these challenges, adaptive concurrency limits and techniques like the TCP congestion control algorithm are suggested for dynamically managing concurrency. Additionally, factors such as payload size and the nature of workloads, whether spiky or consistent, impact concurrency and require careful tuning to maintain desired throughput. Taylor also developed a tool for open-loop testing with ScyllaDB, which complements existing closed-loop testing methods by simulating a constant throughput load, providing insights into real-life system behavior under varying conditions.