benefits of Tair - Tair - Alibaba Cloud Documentation Center

Compared with self-managed Redis databases, ApsaraDB for Redis, or other in-memory databases, Tair performs better in data security, O&M, availability, and kernel optimization.

Comparison between Tair and open source Redis

Item	Tair	Open source Redis
Security protection	Precautionary measures: Supports network isolation over virtual private clouds (VPCs). Supports whitelists for access control. Supports custom accounts and permissions. Operational protection: Supports TLS encryption. Supports transparent data encryption (TDE). Post auditing: Supports audit logs.	Precautionary measures: Requires a self-managed network security system, which is expensive and difficult to build. Comes with risks of data leaks due to the low security of the default Redis access configuration. Does not have an account authentication system. Operational protection: Requires a third-party tool to implement SSL encryption. Post auditing: Does not provide built-in auditing functionality.
Backup and restoration	DRAM-based instances support the data flashback feature that allows you to restore data to a specific point in time. For more information, see Use data flashback to restore data by point in time.	Supports only full data restoration.
O&M	Supports more than ten groups of metrics and a minimum monitoring granularity of microseconds. For more information, see Metrics. Supports alert rules. Supports the change of instance specifications and architecture without service downtime. Specific instance types support imperceptible scaling. Allows you to create instances of different architectures and change instance specifications or architecture. Supports large key analysis based on snapshots, which has high precision and does not cause performance loss. For more information, see Use the offline key analysis feature.	Requires a complex third-party monitoring tool to monitor services. Stops providing services when you change the specifications or architecture. The operation to change the specifications or architecture is complex. Supports large key analysis based on sampling, which can be inaccurate.
Deployment and scaling	Supports elastic scaling and instant instance creation.	Requires the procurement, deployment, and maintenance of hardware, and manual node maintenance.
High availability	Provides the single-zone HA solution. Provides the zone-disaster recovery solution. Uses an independent central module to ensure high availability (HA). This module ensures stable and efficient decision-making and helps prevent split-brain issues.	Requires you to deploy an HA architecture in Sentinel mode in a data center. Requires you to deploy a zone-disaster recovery architecture in Sentinel mode. Uses the Sentinel mode to ensure HA. The Sentinel mode is costly and has a low decision-making efficiency during peak hours. Split-brain issues may occur and affect your services.
Kernel optimization	Provides DRAM-based instances that are equipped with multiple I/O threads. The performance of a DRAM-based instance is three times that of a standard instance with the same specifications. Provides ESSD-based and persistent memory-optimized instances to support large-capacity storage and command-level persistence.	Instances of Redis 6.0 or later support multiple I/O threads to enhance performance. The performance may increase by up to twice, but CPU utilization also becomes high. Uses a storage service such as SSDB or Pika as persistent storage. However, these storage services are not fully compatible with the Redis protocol and allow the management of hot and cold data only at the key level. It is expensive to transfer large keys between memory and disks in these services.
Memory	Provides 100% memory availability. Memory overheads caused by disaster recovery, O&M management, scaling, and data persistence (such as write-time replication of forks) are borne by Alibaba Cloud and are not counted towards instance memory. For example, if you create a 64 GB ApsaraDB for Redis instance, the available memory capacity is 64 GB.	Reserves 25% to 40% of memory resources for disaster recovery, O&M management, and scaling. For example, if you purchase two 64 GB Elastic Compute Service (ECS) instances to create Redis master/replica instances, the available total memory capacity is typically less than 45 GB.

Note

Tair is fully compatible with open source Redis. You can connect to Tair in the same way as you connect to open source Redis. Therefore, you can use any client that is compatible with the Redis protocol to connect to a Tair instance. For more information, see FAQ and Use a client to connect to a Tair instance.

Comparison between Tair and ApsaraDB for Redis

Compared with ApsaraDB for Redis, Tair provides a variety of instance series based on different storage mediums such as DRAM, NVM, and enhanced SSDs (ESSDs) to meet your requirements for low-latency access, persistence, and reduced overall costs. Tair provides you with higher performance, more data modules, and more flexible storage methods. This helps you meet business requirements in different scenarios.

Note

In the following table, ️️️️️️️️️✔️ indicates that this feature is supported, and ❌ indicates that this feature is not supported.
For more information about instance types, see Overview.

Category	Item	Enhanced Edition (Tair)			Community Edition
Category	Item	DRAM-based instance	Persistent memory-optimized instance	ESSD-based instance	Instance that runs Redis 2.8, 4.0, or 5.0	Instance that runs Redis 6.0 or 7.0	Cost-efficient instance that runs Redis 5.0 or 6.0
Baseline performance	Performance benchmark (based on Community Edition)	300%	90%	Read: 50%	Same	120%	120%
	Performance benchmark (based on Community Edition)	300%	90%	Write: 30%	Same	120%	120%
	Maximum number of connections to each data node	30,000	10,000	10,000	10,000	10,000	10,000
	Service capability of a single key (QPS reference value] ①	450,000	130,000	30,000~60,000	140,000	160,000	160,000
	Maximum bandwidth (MB/s)	96~2048	96~2048	128~1024	10~2048	48~2048	96~2048
Specifications	I/O and worker models	Multiple I/O threads + Single worker thread (Real Multi-I/O) ③	Single I/O thread + Single worker thread	Multiple I/O threads + Multiple worker threads (Real Multi-I/O)	Single I/O thread + Single worker thread	Multiple I/O threads + Single worker thread	Multiple I/O threads + Single worker thread
Specifications	Cost per unit (based on Community Edition)	117%	70%	15%~20%	Same	Same	51~67%
Data structures	Basic data structures and supported commands	Different instances support different commands. For more information, see Limits on commands supported by ApsaraDB for Redis Enhanced Edition (Tair).			For more information about the commands that are not supported, see Commands supported by ApsaraDB for Redis Community Edition.
Data structures	Extended data structures of Tair	✔️	️ ️️️✔️️️ (integration with specific Redis modules)	❌	❌	❌	❌
Data persistence	Master-replica replication consistency	Eventual consistency	Eventual consistency	Eventual consistency	Eventual consistency	Eventual consistency	Eventual consistency
	Persistent data consistency ④	Write Back	Write Through	Write Through	Write Back	Write Back	Write Back
	Persistence level	Within seconds	Command-level	Command-level	Within seconds	Within seconds	Within seconds
Security	Audit logs	✔️	✔️	❌	✔️	✔️	✔️
	TLS encryption	✔️	✔️	❌	✔️	✔️	✔️
	TDE	✔️	❌	❌	❌	❌	❌
	Whitelists	✔️	✔️	✔️	✔️	✔️	✔️
Performance analysis	Real-time key statistics	✔️	✔️	✔️	✔️	✔️	✔️
	Query of historical hotkeys	✔️	✔️	❌	✔️	✔️	✔️
	Offline large key analysis	✔️	✔️	❌	✔️	✔️	✔️
Advanced features	Data flashback for data restoration by point in time	✔️	✔️	❌	❌	❌	❌
	Proxy query cache	✔️	❌	❌	❌	❌	❌
	Global Distributed Cache	✔️	❌	❌	❌	❌	❌
	One-way data synchronization by using DTS	✔️	✔️	❌	✔️	✔️	✔️
	Two-way data synchronization by using DTS	✔️	✔️	✔️	❌	❌	❌

The following points illustrate each numeric label:

①: The queries per second (QPS) reference value is measured by a command with a time complexity of O(1). The higher the time complexity, the lower the QPS reference value.
②: This metric is related to the distribution of cold and hot data that is accessed. A higher hit rate on memory indicates that the instance provides the performance closer to that of a Community Edition instance.
③: Different from the multi-threading model of Community Edition 6.0, the Real Multi-I/O model of DRAM-based instances provides fully accelerated I/O threads to support connections and linearly increases throughput.
④: Tair uses the following methods to store data:
- Write Through: writes data directly to disks and returns a success response.
- Write Back: writes data to the cache, returns a success response, and then writes the data to disks.