Umeå universitets logga

Redis has long been the standard for in-memory key–value datastores, but recent licensing changes have prompted many to explore open-source alternatives. This thesis evaluates Valkey, KeyDB, and Garnet as potential successors to Redis, focusing on three core aspects: performance, migration effort, and long-term viability.

Each system was deployed in a Kubernetes-based environment and benchmarked using a modified version of memtier_benchmark. Realistic, Redis-like workloads were used, including Zipfian key-access patterns and varying concurrency levels. In addition to performance metrics like throughput, tail latency, and resource usage, the evaluation also considered migration complexity, such as compatibility with Redis tooling and API, and long-term viability, including development activity, governance, and open-source licensing.

Garnet delivered the highest performance across all metrics, outperforming Redis by a wide margin in throughput and latency while using significantly less memory. However, Garnet is only partially compatible with Redis and does not fully integrate with existing infrastructure, which increases migration complexity. Valkey performed similarly to Redis and scaled better under load, particularly in read-heavy workloads. As a drop-in replacement, Valkey supports the full Redis API and existing workflows, making migration straightforward. KeyDB showed limited performance improvements and no visible development activity during the evaluation period, raising concerns about sustainability.

The results show that Valkey is the most viable long-term alternative to Redis for general use, offering compatibility and operational continuity. Garnet is a strong option for performance-critical applications where full Redis compatibility is not required. 

This study provides a structured, data-driven comparison to support engineers and decision-makers navigating the current Redis ecosystem.