Valkey achieved one million RPS 6 months after forking from Redis

Valkey has introduced significant performance enhancements in its latest version, achieving over 1.19 million requests per second (RPS) through advanced memory access techniques. The blog details how the team offloaded I/O operations to dedicated threads, allowing the main thread to focus on command execution. Profiling revealed that the main thread spent considerable time waiting for external memory access, prompting the implementation of speculative execution and memory access amortization techniques. By interleaving memory access operations, Valkey improved the efficiency of linked list traversals and dictionary lookups, reducing memory access latency. The new approach allows the processor to issue multiple memory accesses in parallel, significantly speeding up operations. For instance, a new interleaved function for summing linked list values reduced execution time from 20.8 seconds to under 2 seconds. Additionally, prefetching memory addresses further optimized performance. The blog also provides a guide for reproducing these performance results on an AWS EC2 instance, detailing hardware setup, server configuration, and benchmark parameters. Valkey 8.0 RC2 is now available for evaluation, showcasing the impact of these optimizations on overall system performance.

- Valkey 8.0 achieves over 1.19 million requests per second.

- Speculative execution and memory access amortization techniques enhance performance.

- Interleaving memory access operations reduces latency significantly.

- A guide is provided for reproducing performance results on AWS EC2.

- Valkey 8.0 RC2 is available for evaluation with new optimizations.