Skip to main content

03 · Network fabric

Pod and storage traffic physically separated. 2 × ToR leaf per rack · MLAG · 25/100G spine. Per-host cabling is in 06 · Physical network topology. · interactive original ↗

Mgmt networkVLAN 100Pod / CNI overlayVXLANLonghorn storageVLAN 200Ceph publicVLAN 210Ceph clusterVLAN 211VM guest VLANs300–399OOB / BMCVLAN 90

Segments​

NetworkVLANSubnetBondPurpose
Mgmt10010.10.100.0/24bond0Harvester/RKE2 control, API VIP, live migration
Pod / CNI—10.42/16 · 10.43/16bond0VXLAN overlay per cluster, rides mgmt underlay
Longhorn20010.10.200.0/24bond1Replica sync + rebuild, isolated from pods
Ceph public21010.10.210.0/24bond1Client I/O: app-cluster CSI ↔ MON/OSD
Ceph cluster21110.10.211.0/24bond1OSD replication + recovery east-west
Guest VLANs300–399per-tenant /24sbond0VM north-south + guest node traffic
OOB / BMC9010.10.90.0/24eth-bmcIPMI/Redfish, air-gapped mgmt switch

Application side · bond0​

Mgmt · VLAN 100​

10.10.100.0/24 — Harvester/RKE2 control, API VIP, live migration. On bond0.

Pod / CNI​

10.42/16 · 10.43/16 — VXLAN overlay per cluster, rides mgmt underlay. Never carries storage replication.

Guest VLANs 300–399​

Per-tenant /24s — VM north-south + guest node traffic. Harvester VLAN networks on bond0.

Storage side · bond1​

Longhorn · VLAN 200​

10.10.200.0/24 — Harvester storage-network setting → dedicated bond1. Replica sync + rebuild isolated from pods.

Ceph public · 210​

10.10.210.0/24 — Client I/O: app-cluster CSI ↔ MON/OSD. Bridged VM NIC on bond1.

Ceph cluster · 211​

10.10.211.0/24 — OSD replication + recovery east-west. Second VM NIC, non-routed.

Out-of-band​

OOB / BMC · VLAN 90​

10.10.90.0/24 — IPMI/Redfish, air-gapped mgmt switch, ACL to jump host only.