Size ElastiCache nodes from dataset size plus key overhead, fragmentation, replicas and reserved-memory headroom.
Four multipliers stack between raw data and node size: per-key overhead, allocator fragmentation, division across shards, and the reserved-memory percentage that ElastiCache holds back so a backup fork or replication buffer cannot trigger an out-of-memory eviction. Replicas add nodes and cost but no usable capacity, because each replica holds a full copy of its shard. The node price is an editable input with a realistic us-east-1 default — confirm current ElastiCache pricing with AWS for your region and node family. A Redis node that runs out of memory starts evicting or, worse, fails a failover mid-fork, so the reserved-memory headroom is not optional padding — it is what makes backups and replica promotion survivable.
ElastiCache Memory
per-node memory = dataset × (1 + overhead) × fragmentation ÷ shards ÷ (1 − reserved-memory percent); nodes = shards × (1 + replicas per shard).
per-node memory = dataset × (1 + overhead) × fragmentation ÷ shards ÷ (1 − reserved-memory percent); nodes = shards × (1 + replicas per shard). Four multipliers stack between raw data and node size: per-key overhead, allocator fragmentation, division across shards, and the reserved-memory percentage that ElastiCache holds back so a backup fork or replication buffer cannot trigger an out-of-memory eviction. Replicas add nodes and cost but no usable capacity, because each replica holds a full copy of its shard. The node price is an editable input with a realistic us-east-1 default — confirm current ElastiCache pricing with AWS for your region and node family.
A Redis node that runs out of memory starts evicting or, worse, fails a failover mid-fork, so the reserved-memory headroom is not optional padding — it is what makes backups and replica promotion survivable.
This calculator takes 9 inputs: Raw dataset size, Key and structure overhead, Memory fragmentation ratio, Reserved memory percent, Shards (node groups), Replicas per shard, Memory per node, Node price, Billed hours per month. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
A replica holds an identical copy of its primary's keyspace, so it adds read throughput and availability but zero extra room for data. Only the shard count multiplies usable memory.