Summarize with AI
Joint reference architecture from VMware by Broadcom and Redis
Moving beyond caching to enterprise-grade real-time data
Teams building fraud detection, streaming analytics, AI agents, and feature stores need fast data access under concurrent load. Running Redis Software on Kubernetes adds questions about worker sizing, network paths, persistence, and recovery.
Broadcom and Redis validated these deployment choices on Dell PowerEdge R770 servers with Intel® Xeon® 6740P processors. The joint reference architecture for enterprise-grade Redis on VMware Cloud Foundation (VCF) with vSphere Kubernetes Service (VKS) combines measured results, sizing guidance, and declarative deployment examples. Platform teams can use Redis Enterprise operator resources to provision clusters and databases through Kubernetes and GitOps workflows.
We ran memtier_benchmark with TLS enabled and separate Redis and client host groups, so benchmark traffic crossed the physical 25 GbE network. The results below reflect the configurations and workloads documented in the reference architecture.
Enterprise-grade Redis, deployed on Kubernetes through the Redis Enterprise operator, supports production real-time data services on VCF and VKS. In joint testing, Redis Software on VCF with VKS delivered:
- Low latency at scale: sub-millisecond combined P99 latency across GET (read) and SET (write) operations on a three-node, six-primary-shard deployment in the reported workloads at pipeline depths 1–2 (one or two outstanding requests per connection). The five-node scale-out configuration, supported by six client worker VMs, delivered approximately 2.0M ops/s at about 1.0–1.5 ms combined P99 latency.
- Online growth: a database with append-only file (AOF) persistence expanded from five to ten primary shards under live traffic in 56 seconds (REDB shard-count change to all shards reporting OK), with zero connection errors recorded by the client.
- Resilience: automatic failover and configurable Redis Database (RDB) snapshots and AOF persistence on vSAN Express Storage Architecture (ESA). In pod-deletion and worker-node drain-and-return tests, Kubernetes restored the Redis pods to Ready and client workloads completed with zero recorded connection errors.
- Multi-tenant efficiency: two concurrent databases on a shared five-node cluster split throughput approximately evenly, with comparable latency under symmetric load.
- Kubernetes-native operations: RedisEnterpriseCluster (REC) and RedisEnterpriseDatabase (REDB) resources let platform teams provision, scale, and manage clusters and databases through declarative Kubernetes workflows.
The blueprint at a glance
Read the full joint reference architecture
Use the reference architecture to select worker sizes, tune networking, configure persistence, and plan online growth. It also covers multi-tenant placement and physical failure domains. Validation covers a single Redis Software cluster; Active-Active multi-site deployment is separate design guidance and was not tested.
Reference architecture: Enterprise-grade Redis on VMware Cloud Foundation 9.1 with vSphere Kubernetes Service
