NFV virtual layer architecture analysis
According to the ETSI reference architecture, the NFVI virtual layer includes two core components: Hypervisor and VIM, and realizes hardware resource abstraction through standard interfaces such as Vi-Ha and Nf-Vi. The key technical features include:
| Architecture level | Functional module | Key technology |
| Compute virtualization | KVM/Qemu | NUMA affinity, CPU core binding |
| Storage virtualization | Cinder | Multi-path access, QoS speed limit |
| Network virtualization | OVS/DPDK | SR-IOV, VxLAN transparent transmission |
Core performance indicators
The standard puts forward strict requirements on the performance of the virtual layer:
- Computing loss: CPU/memory performance loss must be <5% under SPECCPU test
- Network throughput: 10GE network card OVS forwarding throughput > 4Gbps (64-byte packet)
- Fault recovery: VM HA time <30 seconds, node fault recovery <3 minutes
Typical configuration example
An operator's vEPC deployment adopts:
- Computing nodes: 2*Intel Skylake 36 cores/512GB memory
- Storage configuration: Ceph cluster 3 replica strategy
- Network plane: 25GE SR-IOV network card + DPDK acceleration
Reliability design points
The virtual layer needs to achieve 99.999% availability. Key measures include:
- Multi-level redundancy: odd-number deployment of management nodes (3/5-node cluster)
- Fault isolation: network plane division (independent management/business/storage)
- Quick recovery: black box log + WatchDog dual detection mechanism
Implementation suggestions
Note in actual deployment:
- Hypervisor reserved resource control (CPU < 2 cores, memory < 32GB)
- Disable the security group function to improve forwarding performance
- Preferably use virtio-blk disk type