Unlock 3x QPS and microsecond latency with Memorystore for Valkey 9.1

Google Cloud has made Memorystore for Valkey 9.1 generally available, offering up to three times higher queries per second at microsecond latency compared to Memorystore for Redis Cluster. The service builds on Valkey, an open-source fork created after Redis Inc. changed its licensing model in 2024. Valkey is designed to support high-throughput workloads for AI and microservices, where low latency and massive concurrency are critical for backend developers and architects.

Valkey 9.1 introduces a lock-free, multi-queue messaging architecture to improve I/O efficiency in high-throughput datastores. The system replaces static client socket assignment with dynamic work balancing using three queues: a main-to-I/O thread queue for job dispatching, an I/O-to-main thread queue for completed task handling, and dedicated I/O thread queues for memory cleanup and epoll offloading. This architecture reduces CPU waste and prevents thread starvation or hot-spotting.

The update includes automatic CPU-driven activation of background I/O threads when main-thread usage exceeds 30%, followed by dynamic scaling of worker threads based on real-time queue backlog. These improvements aim to maintain performance under heavy load while optimizing resource usage. The changes are designed to keep the main execution loop unblocked, ensuring consistent microsecond latency even as workloads scale.

Memorystore for Valkey 9.1 also introduces new developer capabilities and security enhancements. These include database-level access control within ACL frameworks, topology-aware key scanning with the CLUSTERSCAN command, and centralized management of access control lists at no additional cost. The features address common engineering requests for granular authorization, secure multi-tenancy, and improved observability across database fleets.