9.2 introduces a WireGuard-based SDN fabric option.
Two 9.2 features solve different problems: WireGuard-based SDN fabric connects node networks across sites, while the dynamic load balancer balances VM/CT placement across cluster resources. They are not the same kind of load balancer.
An overlay does not repair a broken underlay; it hides it. Adding WireGuard/BGP on top of packet loss or MTU mismatch makes troubleshooting harder.
Validate underlay IP connectivity, MTU and routing before adding an SDN overlay. Before enabling dynamic balancing, test migration networking, storage constraints and CPU-model compatibility.
Validate underlay IP connectivity, MTU and routing before adding an SDN overlay. Before enabling dynamic balancing, test migration networking, storage constraints and CPU-model compatibility.
Understand Proxmox VE 9.2 WireGuard/BGP SDN fabric and dynamic load balancing with production topology, MTU, routing, migration and capacity safeguards.
9.2 introduces a WireGuard-based SDN fabric option.
Can participate in dynamic routing within the fabric design.
9.2 expands route-map and prefix-list capabilities.
Dynamic load balancing helps placement based on cluster resource balance.
SDN fabric handles network reachability; dynamic placement handles compute placement. HTTP reverse proxies solve application traffic distribution.
| Layer | Example | Problem solved |
|---|---|---|
| Underlay | Physical/VLAN IP | Node-to-node transport |
| Overlay | WireGuard/EVPN | Logical segment reachability |
| Placement | Dynamic load balancer | VM/CT resource placement |
| L7 | HAProxy/Nginx | HTTP request distribution |
IP routes, latency, loss, MTU and clock skew are baseline checks. Cross-site links also require attention to jitter and asymmetric routing.
ip -br addressip routeping -c 20 PEER_IPtracepath PEER_IPchronyc tracking 2>/dev/null || timedatectl statusBlindly using a 1500-byte overlay MTU on a 1500-byte underlay can trigger fragmentation or PMTUD issues. Measure actual path MTU.
tracepath PEER_IPping -M do -s 1472 -c 3 PEER_IPip link showUse 9.2 route-map and prefix-list controls to reduce accidental route leaks. “Redistribute everything” should not be a production default.
Even if placement wants to move a VM, local-only storage, passthrough devices or CPU-model differences may block migration.
| Check | Why |
|---|---|
| Storage | Can target node access disks? |
| CPU model | Live-migration compatibility |
| PCI passthrough | Hardware-bound VM may not move |
| Migration network | Migration time and production impact |
A normal ping test is insufficient. Measure tunnel-peer loss, BGP route loss and compute-node drain scenarios.
Plan 1/10/25G migration networks, VLANs and site-to-site topology for a Proxmox cluster on Eka Sunucu dedicated infrastructure.
Primary documentation and technical references used by this guide.
Continue with related infrastructure and implementation guides.
PVE 9.2 SDN
No. The 9.2 feature targets VM/CT workload placement and cluster resource balance; L7 HTTP load balancing is separate.
Peers need IP-level reachability; whether that is public IP, routed private WAN or another transport depends on topology.
Tunnel encapsulation adds headers. Incorrect path MTU can cause fragmentation, drops or silent application timeouts.
They are typically hardware-bound and not freely movable. Placement policy must account for that constraint.