Tair persistent memory-optimized instances use persistent memory to provide large-capacity in-memory databases that are compatible with open source Redis. A persistent memory-optimized instance persists each operation without the need to rely on traditional disk storage for data persistence. Compared with an ApsaraDB for Redis Community Edition instance, a persistent memory-optimized instance reduces costs by up to 30% and delivers almost the same throughput and latency. This improves the reliability of business data.
The high price and low capacity of memory limit the large-scale use of memory in specific scenarios. Alibaba Cloud began to invest in the research and implementation of persistent memory in 2018. Persistent memory was applied to the core cluster of e-commerce products with remarkably reduced costs during Double 11 that year. The cluster became the first product in China that officially deployed persistent memory in a production environment.
Maturer cloud environments and improved persistent memory technologies help Alibaba Cloud develop a new engine for data persistence implementation. Alibaba Cloud integrates the new engine with Elastic Compute Service (ECS) bare metal instances to introduce Tair persistent memory-optimized instances. These instances replace the traditional volatile memory of Redis with PMEM to significantly reduce the risk of data loss. For more information about ECS bare metal instances, see Overview.
Persistent memory-optimized instances provide not only memory-level access latency and throughput but also data persistence. In addition to reducing costs, persistent memory-optimized instances can simplify the application architecture. The popular architecture that consists of applications, cache, and persistent storage can be simplified to an architecture that consists of applications and persistent memory-optimized instances, as shown in the following figure.
Persistent memory-optimized instances of ApsaraDB for Redis Enhanced Edition (Tair) use persistent memory to provide large-capacity in-memory databases that are compatible with open source Redis. A persistent memory-optimized instance persists each operation without using disks. Compared with an ApsaraDB for Redis Community Edition instance, it reduces costs by up to 30% and delivers almost the same throughput and latency. This helps improve the reliability of business data. Persistent memory-optimized instances are suitable for scenarios that store large amounts of hot and warm data, have high requirements for data persistence and service stability, and require compatibility with Redis.
Integration of multiple data modules
Synchronization mode between the master and replica nodes
In semi-synchronous mode, after the master node processes a request, the system synchronizes logs to the replica node. The master node does not respond to the client until the master node receives a success response from the replica node. This way, data consistency is guaranteed even if a master-replica switchover occurs. For more information, see Modify the synchronization mode of a persistent memory-optimized instance.
Optimization for high specifications
Data loss prevention during power outages
Scenarios that require high performance and reduced costs for processing a large amount of data
Intermediate data computing requires high performance. If you use ApsaraDB for Redis Community Edition for intermediate data computing, the costs are high. Other database types such as HBase cannot meet the performance requirements. Persistent memory-optimized instances not only ensure data persistence but also provide almost the same performance as ApsaraDB for Redis Community Edition instances in terms of throughput and latency. This helps you strike a balance between performance and costs.
Scenarios that have high requirements for data persistence
For gaming services, persistent memory-optimized instances are used for data storage. Compared with a Redis and MySQL-based architecture, persistent memory-optimized instances provide a simpler architecture, and higher performance, cost-effectiveness, and data reliability.