Skip to main content

Overview

Xloud Networking follows a distributed agent model. A central API and database tier manages resource state, while per-node agents program the virtual switching fabric in real time. Understanding this architecture helps administrators diagnose failures, plan capacity, and maintain the networking plane across the cluster.
Administrator Access Required — This operation requires the admin role. Contact your Xloud administrator if you do not have sufficient permissions.

Architecture Diagram


Agent Components

AgentDefault PortRole
Networking API9696RESTful endpoint for all network resource management
L2 AgentN/A (message bus)Programs virtual switching and port bindings on each compute node
L3 AgentN/A (message bus)Manages routers, NAT rules, and floating IP translation
DHCP Agent67/68 (DHCP)Provides IP address assignment for tenant subnets
Metadata Agent80 (internal proxy)Forwards instance metadata requests from the network namespace
RPC Message Bus5672 (AMQP)Carries control messages between the API server and distributed agents

Request Flow: Create Network and Launch Instance


Data Plane: Virtual Switching

Each compute node runs an L2 agent that programs the local virtual switch to enforce:
  • Port bindings — connect instance virtual NICs to the correct network segment
  • VLAN or VXLAN segmentation — isolate tenant traffic from other projects
  • Security group rules — iptables/nftables rules enforced per port
  • Anti-spoofing — MAC and IP address binding prevents address impersonation
The specific virtual switch backend (Linux bridge, Open vSwitch, or OVN) is configured during cluster deployment via XDeploy. The choice of backend affects performance characteristics and advanced features like DVR and hardware offload.

High Availability Considerations

ComponentHA MechanismImpact of Failure
Networking APIMultiple API instances behind HAProxySingle instance loss: no impact
DatabaseGalera cluster (3+ nodes)Partial loss: degraded writes
Message BusRabbitMQ cluster (3+ nodes)Partial loss: agent messaging delayed
L2 AgentOne per compute nodeAgent loss: no new port bindings on that host
L3 AgentVRRP failover (HA routers)Active agent loss: standby takes over
DHCP AgentMultiple agents per networkAgent loss: secondary agent serves leases

Next Steps

Network Agent Management

Monitor agent health and manage agent lifecycle

Provider Networks

Configure physical network mappings and segmentation types

L3 Router Configuration

Enable HA and distributed routing for production deployments

DHCP Configuration

Manage DHCP agents and subnet assignments