Data tiering

Clusters that use a node type from the r6gd family have their data tiered between memory and local SSD (solid state drives) storage. Data tiering provides a new price-performance option for Valkey and Redis OSS workloads by utilizing lower-cost solid state drives (SSDs) in each cluster node in addition to storing data in memory. Similar to other node types, the data written to r6gd nodes is durably stored in a multi-AZ transaction log. Data tiering is ideal for workloads that access up to 20 percent of their overall dataset regularly, and for applications that can tolerate additional latency when accessing data on SSD.

On clusters with data tiering, MemoryDB monitors the last access time of every item it stores. When available memory (DRAM) is fully consumed, MemoryDB uses a least-recently used (LRU) algorithm to automatically move infrequently accessed items from memory to SSD. When data on SSD is subsequently accessed, MemoryDB automatically and asynchronously moves it back to memory before processing the request. If you have a workload that accesses only a subset of its data regularly, data tiering is an optimal way to scale your capacity cost-effectively.

Note that when using data tiering, keys themselves always remain in memory, while the LRU governs the placement of values on memory vs. disk. In general, we recommend that your key sizes are smaller than your value sizes when using data tiering.

Data tiering is designed to have minimal performance impact to application workloads. For example, assuming 500-byte String values, you can typically expect an additional 450 microseconds of latency for read requests to data stored on SSD compared to read requests to data in memory.

With the largest data tiering node size (db.r6gd.8xlarge), you can store up to ~500 TBs in a single 500-node cluster (250 TB when using 1 read replica). For Data tiering, MemoryDB reserves 19% of (DRAM) memory per node for non-data use. Data tiering is compatible with all Valkey and Redis OSS commands and data structures supported in MemoryDB. You don't need any client-side changes to use this feature.