Help Center/ Distributed Cache Service/ Service Overview/ Selecting DCS Products
Updated on 2025-11-03 GMT+08:00
View PDF

Selecting DCS Products

Select a DCS Redis instance based on your service scenarios, performance requirements, and budget. This document provides references for selecting product editions, instance types, and instance specifications.

Selecting a Product Edition

Table 1 Product editions

Preference

Suggestion

Analysis

Performance and enterprise-class capabilities

Enterprise (Performance)

(Enterprise (Memory))
  • Performance: multithreading and multi-core scheduling, higher performance at a single node, and more stable access latency in the case of high concurrency
  • Enterprise-level capabilities: data flashback
  • Rich modules: JSON and Bitmap

Balanced performance and cost

Basic edition
  • Performance: slightly higher than that of open-source self-built Redis, but lower than that of the enterprise (memory) edition.
  • High availability: Master/Standby replication and automatic failover are available.
  • High elasticity: Vertical and horizontal scaling satisfies capacity and performance requirements.

Cost-effectiveness

Enterprise (storage) edition

(Enterprise (capacity) edition)
  • Performance: equivalent to that of open-source self-built Redis.
  • Low cost: With cold/hot separation, the SSD stores full data and the memory stores hot data. Therefore, the cost is the lowest and the performance is basically the same when the capacity is the same.

Ultra-large bandwidth

Basic edition Redis Cluster instance
  • Assured bandwidth: Compared with Proxy Cluster instances whose maximum bandwidth is 10,000 Mbit/s, bandwidth of a Redis Cluster instance can be increased as the number of shards increases.
  • Elastic bandwidth: On each shard, the bandwidth can be increased to 2,048 Mbit/s.
  • Enterprise (performance) and enterprise (storage) editions are in upgrades. They will become enterprise (memory) and enterprise (capacity) editions. To protect customer rights and reduce O&M complexity, enterprise features will be available later. Enterprise edition instances are currently in restricted sales.
  • For details about the differences between DCS editions, see Comparing Redis Instance Types.

Selecting an Instance Type

Table 2 Instance types and replica quantities

Category

Description

Scenarios

Single-node

Single replica (one node). Data persistence is unavailable. The instance is inaccessible and data will be lost when the node is faulty.

Only for development and test scenarios. Do not use it in a production environment. There is no Service Level Agreement (SLA) assurance.

Master/Standby

Two replicas (one master node and one standby node). Data persistence is available. Node faults can be rectified through master/standby switchovers, ensuring access.

Production services whose capacity is small, performance is stable at the 100,000 QPS level, and expansion is not required.

Proxy Cluster

Two replicas (one master node and one standby node) by default. Forward proxies shield backend network topology changes, reducing client access complexity.

Production services whose data capacity is large and concurrency is in the tens of millions.

Redis Cluster

Starting with two replicas (one master node and one standby node) is recommended. Full compatibility with open-source Redis.

Production services whose data capacity is large and concurrency is in the tens of millions.

Read/Write splitting

At least three replicas (one master node and at least two standby nodes). Based on the architecture of a master/standby instance, forward proxies schedule read and write requests to the master and standby nodes respectively.

Production services whose data capacity is small, write performance is at the 100,000 QPS level, and read performance is at the 100,000s QPS level.

Selecting an Instance Flavor

DCS Instance Specifications compares the instance specifications supported by different instance types. An instance flavor can be estimated with read/write concurrency, data volume, and maximum bandwidth. Reserve at least 20% more than the estimated requirement can reduce complexity in O&M changes later.

The following provides an example of estimating the flavor of a basic edition instance:

  • When read/write performance is at the 100,000 QPS level, select master/standby.
  • When write performance is at the 100,000 QPS level and read concurrency is at the 100,000s QPS level, read/write splitting is recommended.
  • When read and write performance is at the 100,000s QPS level, Proxy Cluster or Redis Cluster is required. Select a shard quantity based on performance requirements.
  • Reserve at least 20% more than the required specifications.