Hybrid-storage instances are retired. For details, see End-of-sale for Tair hybrid-storage instances. Switch to persistent memory-optimized instances instead.
Tair hybrid-storage instances are developed by Alibaba Cloud and are compatible with the Redis protocol. They store hot data in memory and all data on disk simultaneously—delivering high read and write performance without the constraints of memory capacity. Unlike Redis Open-Source Edition instances, which are memory-only, hybrid-storage instances let you manage large datasets cost-effectively while keeping frequently accessed data fast.
How it works
Figure 1. Hybrid-storage instance architecture 
Hybrid-storage instances use two storage layers that work together:
In-memory storage: Keys and values of hot data are stored in memory. The information of all keys is also cached in memory, so key existence checks are always fast regardless of where the value lives.
Disk storage: All keys and values are persisted to disk. Redis data structures such as hashes are stored on disk in a specific format.
This two-layer design gives your working set memory-speed access while the full dataset is safely persisted—without requiring enough memory to hold everything at once.
Use cases
Workloads that benefit:
| Use case | How hybrid storage helps |
|---|---|
| Livestreaming | Popular live channels generate bursts of hot data. Store popular channel data in memory and less popular channels on disk to make the most of limited memory. |
| E-commerce | Product catalogs are large, but only a fraction of items are actively browsed. Keep popular product data in memory and the full catalog on disk. |
| Online education | Course libraries and question banks are large, but only current courses and active question sets are accessed frequently. Store the full content on disk and hot courses in memory. |
Cost examples
Hybrid-storage instances can reduce storage costs by more than 50% compared to an equivalent all-memory cluster—particularly when most of your data is cold.
Example 1: A native Redis cluster stores 100 GB of data. Peak queries per second (QPS) stays below 20,000, and 80% of the data is rarely accessed.
A hybrid-storage instance with 32 GB memory and 128 GB storage capacity handles the same workload, saving approximately 70 GB of memory resources and cutting storage costs by more than 50%.
Example 2: An on-premises Pika instance is used to reduce the storage costs of a native Redis deployment. The total size of the data is roughly 400 GB, with only about 10% accessed frequently. Operations and maintenance (O&M) overhead for the cluster is high.
A hybrid-storage instance with 64 GB memory and 512 GB storage capacity handles this workload with reduced O&M costs and built-in high availability.