Designing a CPU-Efficient Redis Cluster Topology

Redis is a popular in-memory data store that has become an essential component of many modern applications. With its high performance, scalability, and reliability features, Redis has emerged as a top choice for caching, session management, and other use cases. In this article, we'll explore the deployment topology of Redis Cluster, specifically focusing on the master-replica approach utilizing all the cores on the vms, leveraging the single threaded behaviour of redis.

What Is a Redis Cluster

A Redis Cluster is a distributed deployment that shards your dataset across multiple Redis nodes. It automatically handles data partitioning and replication, ensuring both high availability and horizontal scalability.

Each cluster node manages a subset of hash slots, allowing the system to distribute data and load efficiently. When combined with replicas, Redis Cluster provides fault tolerance and performance benefits ideal for high-traffic workloads.

Typical Master-Replica Deployment

In a standard Redis deployment, a master node handles all write operations while replica nodes replicate data and handle read operations.

For example, a 3-node setup might consist of 1 master and 2 replicas.

Pros:

• Availability: If the master fails, one of the replicas is automatically promoted
• Fault Tolerance: Adding more replicas increases redundancy
• Load Distribution: Reads and writes are separated across nodes for efficiency

Cons:

• Limited scalability for writes, since only one master handles them
• Under utilization of CPU resources, as Redis uses a single thread per process

Multi-Master Cluster Deployment

To overcome the write scalability issue, Redis supports multiple masters, each with their own set of replicas.

Let’s consider a 3-master and 2-replica setup, totaling 9 VMs (2 vcpu each).

Pros:

• Sharding: Each master handles a unique keyspace segment — distributing load effectively
• High Write Throughput: Multiple masters process writes concurrently
• Fault Isolation: A single master failure impacts only a subset of keys

Cons:

• Infrastructure Cost: Requires significantly more VMs or instances
• Operational Complexity: Managing hash slot rebalancing and failover increases overhead
• Potential Imbalance: Uneven key distribution can cause hotspots

Optimized Deployment: CPU-Aware Containerized Redis Cluster

To reduce infrastructure cost while maintaining performance, we can consolidate Redis masters and replicas on fewer VMs by using Docker Swarm and CPU pinning.

Deployment Strategy

• Three VMs, each with 4 CPU cores
• Docker Swarm configured across all three VMs
• On each VM, run three Redis containers:
  • One master container (unique per VM)
  • Two replica containers — each replicating masters from the other VMs

In this topology:

Each Redis process (master or replica) runs on a dedicated CPU core.

Redis’ single-threaded design ensures one vCPU per instance provides optimal performance.

Each VM uses:

• 1 core for its master
• 2 cores for replicas of the other masters
• 1 core for system operations

Advantages:

• CPU Efficiency: Fully utilizes available cores without over provisioning
• Cost Optimization: Achieves multi-master performance using only 3 VMs instead of 9
• Simplified Management: Fewer VMs to monitor, patch, and secure
• Same High Availability: Replication ensures data redundancy across hosts

Implementation Specific:

• Create a Docker swarm with those 3 vms
• Create a Docker overlay network. (--attachable flag, so that standalone containers can attach to this network)
• Open following ports on all the nodes: Redis port to serve client (6379) and cluster bus port (16379, 16380, 16381)
• Create the Docker Redis container on all the vms, using the overlay network and exposing the ports, on all the 3 vms

Once the docker Redis container are up, we can create Redis cluster using below command from any Redis container.

docker exec -it redis-master1 redis-cli --cluster create <<redis-master1-ip>>:6379 <<redis-master2-ip>>:6379 <<redis-master3-ip>>:6379 \
<<redis-master1-ip>>:6380 <<redis-master2-ip>>:6380 <<redis-master3-ip>>:6380 \
<<redis-master1-ip>>:6381 <<redis-master2-ip>>:6381 <<redis-master3-ip>>:6381  \
--cluster-replicas 2

Once you run above command, you can confirm cluster creation using following command:
docker exec -it redis-master1 redis-cli cluster nodes

Testing

We were able to validate using a standard python (locust) script, that only a single core is utilized when Redis process (container) is deployed on a 2 core vm.

While running the same locust python script (with set/get to varied data-structures), we observed that Redis master were handling the writes and only used one core. The load was evenly distributed across all 3 masters.