Overview of Alibaba Cloud Redis Architecture and Optimizations

This article compiles a sharing from Alibaba student Xia Zhou , detailing the development and current status of Alibaba Cloud Redis.

Redis Introduction

Redis: Remote Dictionary Server, a key‑value storage system. Advantages include:

Ease of use: rich data‑structure support and modules.

High performance: efficient data‑structure design, fully in‑memory operations.

Reliability: master‑slave synchronization and persistence.

Alibaba Cloud Redis Architecture

Overall Architecture

Single‑node → single‑machine master‑slave → cluster → read‑write separation → intra‑city disaster recovery → inter‑city multi‑active.

The Alibaba Cloud overall architecture includes six supporting systems:

HA control system – monitors Redis instance health.

Log collection system – gathers slow‑query, access, and other logs.

Monitoring system – collects performance metrics such as basic info, keys, strings, etc.

Online migration system – rebuilds instances from backup files when the underlying physical machine fails.

Backup system – backs up Redis instances to OSS, supporting user‑defined schedules and a 7‑day retention.

Task control system – handles creation, configuration changes, backups, and tracks task execution and errors.

Dual‑Replica Architecture

Applicable scenarios: pure cache, data persistence.

Performance: 80‑100k QPS.

Connection path: internal SLB → Redis.

Access method: compatible with all open‑source clients.

SLA: 2 replicas, high‑availability, sub‑second failover.

Cluster Dual‑Replica Architecture

Applicable scenarios: large data volume, high performance requirements.

Performance: 1M QPS.

Connection path: internal SLB → Redis.

Access method: compatible with all open‑source clients.

SLA: sharded 2‑replica high‑availability.

Read‑Write Separation Architecture

Applicable scenarios: read‑heavy, write‑light, large keys, no strong consistency requirement.

Advantages: linear scaling of read/write capacity, supports all commands, transparent to users.

Intra‑City Disaster Recovery

High data availability; hot‑standby across two data centers ensures seamless failover; after recovery, incremental sync via Binlog prevents traffic spikes.

Inter‑City Multi‑Active Disaster Recovery

Applicable scenarios: multi‑region read/write, proximity‑based access.

Advantages: tolerates N‑1 data‑center failures, eventual consistency, cross‑region disaster recovery.

Hybrid Storage

Applicable scenarios: massive data, moderate performance, video streaming, e‑commerce.

Advantages: 100% Redis compatibility, hot‑cold data separation, high cost‑effectiveness, asynchronous write‑back of cold data to RocksDB, single instance supports TB‑scale storage.

Alibaba Cloud Redis Kernel Optimizations

Compatibility with Memcache protocol.

High‑availability probing.

AOF Binlog support.

Persistence system enhancements.

Security encryption.

I/O and connection optimizations.

Persistence System Refactor

Retains historical AOF logs, removes AOF rewrite, expands AOF log information, and adopts a new data organization: full RDB + historical AOF.

Asynchronous AOF Write Improvements

Added biowrite mode.

Lightweight lock queue.

Dedicated BIO thread for write operations.

Reduces impact of slow writes on the main thread, boosting performance.

Memcache Protocol Support

Supports both text and binary protocols, fully compatible.

Reuses Redis String objects to store memcached data.

Leverages native Redis synchronization and persistence mechanisms.

New version adds scanning, backup, and richer statistics.

Hybrid Storage Optimizations

All keys and hot data stay in memory.

Cold data stored in RocksDB.

Hot‑data performance matches native Redis.

100% Redis compatibility.

Cold‑data write‑back to RocksDB is asynchronous, not blocking the main thread.

Single instance can handle TB‑scale data.

Network I/O multi‑threading optimizations further improve throughput.