The logical architecture presents aNET as an overlay network fabric built on top of the physical network underlay. It centralizes the control and management plane for all virtual networking elements while distributing the data plane across each hypervisor host. This model creates a unified network domain where policies are defined once and enforced everywhere, providing consistent connectivity and security for virtual machines regardless of their physical location within the data center.
Functional Architecture of aNET
aNET's functional architecture comprises specialized modules that handle distinct aspects of network operations. The architecture cleanly separates the control plane, responsible for topology management and policy distribution, from the data plane, which handles actual packet forwarding. These modules are deployed across the HCI environment, with control plane components typically clustered on management nodes and data plane components running on every host to ensure both centralized management and localized, high-performance packet processing.
High Availability Design of aNET
High availability is engineered into each layer of aNET's architecture. The Control and Management Plane employs a clustered design for automatic failover. The Cluster Communication Plane maintains redundant pathways and heartbeats to preserve cluster cohesion. The Message Bus ensures reliable inter-service communication through redundant brokers and persistent queues, guaranteeing that critical network control messages are never lost, even during component failures.
Reliable Network Planes of aNET
aNET establishes four logically isolated network planes: management, business, storage, and the VXLAN data plane. Each plane serves a specific operational purpose, and their logical isolation ensures that a failure in one plane does not impact the others. These planes can be deployed on separate physical networks or share infrastructure with quality of service guarantees, providing flexibility in network design while maintaining robust fault containment.
High-Performance Architecture of aNET
The architecture incorporates multiple optimizations to maximize throughput and minimize latency. It utilizes technologies like DPDK for kernel-bypass packet processing, multiple queues with interrupt binding for NICs, and NUMA node affinity for optimal resource localization. These optimizations collectively ensure the virtualized network can support demanding workloads with near-physical performance characteristics.