Redis Flex and Auto Tiering
Redis Flex and Auto Tiering enable your data to span both RAM and dedicated flash memory.
| Redis Enterprise Software | Redis Cloud |
|---|
Redis Flex and Auto Tiering in Redis Enterprise Software enable databases to use solid state drives (SSDs) to extend beyond DRAM capacity. Developers can build applications that require large datasets using the same Redis API. Using SSDs can significantly reduce the infrastructure costs compared to only DRAM deployments.
Frequently used data, called hot data, belongs in the fastest memory level to deliver a real-time user experience. Data that is accessed less frequently, called warm data, can be kept in a slightly slower memory tier. Redis Flex maintains hot data in DRAM, keeps warm data in SSDs, and transfers data between tiers automatically.
Redis Flex is based on a high-performance storage engine (Speedb) that manages the complexity of using SSDs and DRAM as the total available memory for databases in a Redis Enterprise cluster. This implementation offers a performance boost of up to 10k operations per second per core of the database, doubling the performance of Redis on Flash.
Just like all-RAM databases, Redis Flex databases are compatible with existing Redis applications.
Redis Flex is also supported on Redis Cloud and Redis Enterprise Software for Kubernetes.
Redis Flex versus Auto Tiering
Redis Flex is the enhanced successor to Auto Tiering, generally available as of Redis Enterprise Software version 8.0.2.
Redis database versions support Redis Flex and Auto Tiering as follows:
| Redis database version | Redis Flex | Auto Tiering |
|---|---|---|
| 8.0 and later | ✅ | ❌ |
| 7.4 | ✅ | ✅ |
| 7.2 and earlier | ❌ | ✅ |
Redis Flex requires the Speedb driver, while Auto Tiering can use either RocksDB or Speedb. See Manage Auto Tiering storage engine for more information.
Use cases
The benefits associated with Redis Flex are dependent on the use case.
Redis Flex is ideal when your:
- working set is significantly smaller than your dataset (high RAM hit rate)
- average key size is smaller than average value size (all key names are stored in RAM)
- most recent data is the most frequently used (high RAM hit rate)
Redis Flex is not recommended for:
- Long key names (all key names are stored in RAM)
- Broad access patterns (any value could be pulled into RAM)
- Large working sets (working set is stored in RAM)
- Frequently moved data (moving to and from RAM too often can impact performance)
Redis Flex is not intended to be used for persistent storage. Redis Enterprise Software database persistent and ephemeral storage should be on different disks, either local or attached.
Where is my data?
When using Redis Flex, RAM storage holds:
- All keys (names)
- Key indexes
- Dictionaries
- Hot data (working set)
All data is accessed through RAM. If a value in flash memory is accessed, it becomes part of the working set and is moved to RAM. These values are referred to as "hot data".
Inactive or infrequently accessed data is referred to as "warm data" and stored in flash memory. When more space is needed in RAM, warm data is moved from RAM to flash storage.
RAM to Flash ratio
Redis Enterprise Software allows you to configure and tune the ratio of RAM-to-Flash for each database independently, optimizing performance for your specific use case. While this is an online operation requiring no downtime for your database, it is recommended to perform it during maintenance windows as data might move between tiers (RAM <-> Flash).
The RAM limit cannot be smaller than 10% of the total memory. We recommend you keep at least 20% of all values in RAM. Do not set the RAM limit to 100%.
Flash memory
Implementing Redis Flex requires pre planning around memory and sizing. Considerations and requirements for Redis Flex include:
- Flash memory must be locally attached. Using network-attached storage (NAS), storage area networks (SAN), or solutions such as AWS Elastic Block Storage (EBS) is not supported.
- Flash memory must be dedicated to Redis Flex and not shared with other parts of the database, such as durability, binaries, or persistence.
- For the best performance, the SSDs should be NVMe based, but SATA can also be used.
- The available flash space must be greater than or equal to the total database size (RAM+Flash). The extra space accounts for write buffers and write amplification.
After these requirements are met, you can create and manage both Redis Flex databases and all-RAM databases in the same cluster.
When you begin planning the deployment of a Redis Flex database in production, we recommend working closely with the Redis technical team for sizing and performance tuning.
Cloud environments
When running in a cloud environment:
- Flash memory is on the ephemeral SSDs of the cloud instance (for example the local NVMe of AWS i4i instances and Azure Lsv2 and Lsv3 series).
- Persistent database storage needs to be network attached (for example, AWS EBS for AWS).
On-premises environments
When you begin planning the deployment of Redis Flex in production, we recommend working closely with the Redis technical team for sizing and performance tuning.
On-premises environments support more deployment options than other environments such as:
- Using Redis Stack features:
Size limits for keys and values
Redis Flex databases cannot store keys or values larger than 4GB in flash storage.
Keys or values larger than 4GB will be stored in RAM only, and warnings will appear in the Redis logs similar to:
# WARNING: key too big for disk driver, size: 4703717276, key: subactinfo:htable