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

Google Cloud has made Memorystore for Valkey 9.1 generally available, promoting it as delivering up to 3x higher queries per second with microsecond‑level latency compared with its Redis Cluster service. The release replaces the previous static round‑robin thread polling with a lock‑free, multi‑queue messaging system that moves work dynamically between the main thread and I/O workers. A two‑phase scaling engine now starts an additional I/O thread when main‑thread CPU usage exceeds 30 percent and adjusts the pool of workers based on real‑time queue depth. The architectural changes aim to keep the main execution loop unblocked, which is critical for workloads that need massive throughput while preserving ultra‑low latency, such as AI back‑ends and high‑scale microservice environments. By reducing cross‑thread CPU waste and enabling work‑stealing, developers can handle millions of concurrent users without the latency spikes typical of static thread models. The new service also adds granular ACLs and a cluster‑wide scanning command, broadening its appeal for secure, multi‑tenant deployments. Google Cloud states the performance claims, but independent benchmarks have not been published, leaving the exact magnitude of the latency improvement open to verification. It is unclear how the dynamic scaling behaves under extreme load spikes beyond the documented 30 percent CPU trigger. Additionally, while the CLUSTERSCAN command is described as topology‑aware, real‑world reliability during rapid slot migrations remains to be demonstrated.