Size Memorystore for Redis from your dataset plus fragmentation headroom, then price replicas and high availability.
Redis needs materially more memory than the dataset it holds, because fragmentation and copy-on-write during a snapshot both consume headroom — sizing to exactly the dataset is how instances start evicting keys under load. Every copy is billed, so Standard tier plus one read replica is three times the capacity charge of a Basic instance. A cache that starts evicting keys silently shifts load back onto your database, which looks like a database problem, and the fix is memory headroom rather than tuning. Confirm the GB-hour rate for your capacity tier and region in the Google Cloud pricing calculator, since the rate steps down as instances get larger.
Memorystore Capacity
provisioned GB = ceil(dataset ÷ (1 − headroom) ÷ step) × step; billable GB = provisioned × copies, where Standard counts 2 copies plus one per read replica; cost = billable GB × GB-hour price × hours.
provisioned GB = ceil(dataset ÷ (1 − headroom) ÷ step) × step; billable GB = provisioned × copies, where Standard counts 2 copies plus one per read replica; cost = billable GB × GB-hour price × hours. Redis needs materially more memory than the dataset it holds, because fragmentation and copy-on-write during a snapshot both consume headroom — sizing to exactly the dataset is how instances start evicting keys under load. Every copy is billed, so Standard tier plus one read replica is three times the capacity charge of a Basic instance.
A cache that starts evicting keys silently shifts load back onto your database, which looks like a database problem, and the fix is memory headroom rather than tuning. Confirm the GB-hour rate for your capacity tier and region in the Google Cloud pricing calculator, since the rate steps down as instances get larger.
This calculator takes 10 inputs: Working dataset size, Headroom for fragmentation and replication buffers, Service tier, Additional read replicas, Price per GB-hour, Hours in the billing month, Capacity step, Network throughput per GB, Network throughput ceiling, Average value size. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
No. Read replicas scale read throughput; automatic failover comes from the Standard tier. A Basic instance with no replica loses its entire dataset on a node failure and every request falls through to the origin datastore until the cache is warm again.