June 2026 Summaries

The comparison between Dragonfly and Valkey 9.0 on AWS Graviton instances reveals significant advancements in Valkey's I/O throughput, especially in read-heavy workloads on smaller instances, narrowing the performance gap established with Valkey 8.0.2. Despite these improvements, Dragonfly's architecture still leads in write throughput, vertical scaling, tail latency, and memory efficiency, with Dragonfly achieving up to 6.6 million QPS in a single process setup on c7gn.metal instances. The tests were conducted with identical hardware and network configurations to ensure unbiased results, and they highlight the strengths of Dragonfly's thread-per-core design compared to Valkey's single-threaded command execution model. While Valkey 9.0 shows progress in certain areas, particularly for read-dominant and tail-tolerant workloads, Dragonfly continues to excel in overall performance metrics, offering a scalable solution without the need for clustering.

Dragonfly SSD Data Tiering, now generally available, extends Dragonfly's in-memory data store capabilities by utilizing local NVMe SSDs to manage datasets larger than the available DRAM without requiring resharding or code changes. This feature maintains hot data in RAM for sub-millisecond latency while offloading warm and cold data to SSDs, ensuring keys and metadata remain in memory for efficient operations. Unlike Redis and Valkey, Dragonfly provides a Redis-compatible, source-available store with transparent SSD tiering, enhancing scalability and cost-efficiency without API changes. Its architecture mitigates the latency typical of disk I/O through a fiber-based execution model, non-blocking I/O with Linux io_uring, and a three-state data model that efficiently manages hot, cold, and cooled data states. Meesho, an Indian e-commerce platform, experienced significant performance improvements and cost savings using Dragonfly's tiering system compared to previous configurations. The tiering solution, optimized for larger String values, offers a seamless setup and is currently in beta for Dragonfly Cloud, promising scalable vertical growth without extensive infrastructure changes.