Tair (Redis® OSS-Compatible):Comparison between Tair (Redis OSS-compatible) and self-managed Redis

Dimension

Tair (Redis OSS-compatible)

Self-managed Redis

Security

Network isolation: Uses virtual private clouds (VPCs) to isolate traffic at the network level.

Access control: Whitelist-based access control (see Configure whitelists) and custom database accounts with granular permissions (see Create and manage database accounts).

Encryption in transit: Native TLS encryption.

Audit logs: Supports audit logs (see Query audit logs).

Network isolation: Requires a custom network security system, which is costly to build and maintain. Default Redis access configuration has security vulnerabilities that can expose data.

Access control: No built-in account authentication system.

Encryption in transit: Requires a third-party tool to implement SSL encryption.

Audit logs: No auditing feature.

Backup and restore

Tair (Enterprise Edition) DRAM-based instances support data flashback — restore data to any specific point in time. See Use data flashback to restore data by point in time.

Full data restoration only — no point-in-time recovery.

O&M

• Over ten groups of monitoring metrics with a minimum monitoring interval of 5 seconds. See Metrics.

• Configurable alert rules. See Alert settings.

• Supports multiple instance architectures; change specifications or architecture as needed.

• Large key analysis based on snapshots — high accuracy, no performance impact. See Use the offline key analysis feature.

• Requires third-party tools for monitoring.

• Changing instance specifications or architecture requires a service interruption and involves complex manual steps.

• Large key analysis based on sampling — results can be inaccurate.

Deployment and scaling

Elastic scaling and instant instance creation — no hardware to provision or manage.

Requires hardware procurement, data center setup, and node deployment. Adding nodes requires manually managing node relationships.

High availability (HA)

• Single-zone HA and zone-disaster recovery solutions available.

• Uses an independent central module for failover decisions — stable, efficient, and resistant to split-brain issues.

• HA in Sentinel mode within a data center, or zone-disaster recovery in Sentinel mode.

• Sentinel mode has high operational cost, slower decision-making under peak load, and is susceptible to split-brain issues.

Kernel optimization

• Tair (Enterprise Edition) DRAM-based instances use a multi-threaded model, delivering 3× the performance of a standard instance at the same specifications.

• ESSD/SSD-based and persistent memory-optimized instances support large-capacity storage and command-level persistence.

• Redis 6.0 and later support multiple I/O threads, which can increase throughput by up to 2× but also raises CPU utilization significantly.

• Persistent storage relies on third-party solutions such as SSDB or Pika, which are not fully compatible with the Redis protocol and support hot/cold data tiering at the key level only. Transferring large keys between memory and disk is expensive.

Memory utilization

100% of allocated capacity is available to your application. Memory overhead for disaster recovery, O&M, scaling, and data persistence (including fork-based write-time replication) is covered by Alibaba Cloud.

Example: A 64 GB instance provides 64 GB of usable memory.

25%–40% of total memory is reserved for disaster recovery, O&M, and scaling operations.

Example: Two 64 GB Elastic Compute Service (ECS) instances configured as a master/replica pair typically provide less than 45 GB of usable memory in total.