# Aviz Networks details tenant network isolation with vCluster

URL: https://decisioninsights.ai/aviz-networks-details-tenant-network-isolation-with-vcluster/

Aviz Networks and vCluster have integrated products that automate tenant network isolation for multi-tenant AI Cloud providers. The workflow moves GPU servers between tenant networks as compute is reassigned, replacing manual switch changes that can expose traffic.Network Changes Lag Compute ReassignmentMulti-tenant AI Cloud providers and AI factories increasingly reassign GPU servers on demand rather than under long-term bare-metal contracts. Compute can be handed to a new tenant in minutes, but tenant network changes such as moving switch ports, VLANs, or VNIs are typically made manually through the CLI. A server left on a prior tenant's segment can expose that tenant's traffic, and the article identifies configuration drift and mixed switch fabrics as added burdens.Joint ArchitecturevCluster coordinates the tenant workflow and decides which server is assigned to which tenant, while vMetal provisions the servers through one API with drivers such as Metal3. When vCluster assigns or releases a server, it calls Aviz ONES, which adds or removes the server's switch ports from the tenant's VXLAN or VLAN segment on the provider's switch fabric.Each tenant receives a dedicated control plane and Private Nodes, and vCluster assigns tenant IP addresses from the subnet the provider configures. Aviz ONES also provides an isolated holding network for unassigned servers and VPC peering for shared services such as storage, DNS, or registries.Server Lifecycle WorkflowThe workflow contains four stages: vMetal provisions a server on a provisioning network separate from tenant networks, vCluster assigns it to a tenant, and Aviz ONES attaches it to the tenant segment. For tenant clusters the server joins as a Private Node, and on release vCluster detaches it from the tenant network before vMetal cleans it for the next tenant.According to the article, isolation depends on the provider's fabric configuration, so providers validate it before running tenant workloads. The integration is available in vCluster v5.0 through the Metal3 driver in vMetal, with a vCluster license that includes the Aviz integration.Supported Platforms and DeploymentsAviz ONES supports one tenant model across SONiC, NVIDIA Cumulus Linux and InfiniBand, Arista EOS, and Cisco NX-OS, and includes fabric telemetry with optional export to Splunk, ELK, or ServiceNow. vCluster offers bare metal, Kubernetes, Slurm, and inference cluster types across GPU generations.The article states that vCluster runs in production across more than 100K GPUs at 50-plus AI Clouds and Fortune 500 companies. It cites QumulusAI spinning up isolated Kubernetes environments in one minute and 35x faster customer onboarding at Deutsche Telekom.The integration gives AI Cloud providers and AI factories a tenant boundary extending from the control plane to the switch port, so each tenant runs on its own virtual network on the provider's own hardware. This Blog Summary is a fact-based summary of the vendor blog and is intended for enterprise IT and network decision-makers.
