HA VPN Explained: The 99.99% SLA, Peer Topologies and Scaling Throughput

GCP Study Hub
November 8, 2026

HA VPN is the Cloud VPN option that Google covers with a 99.99% availability SLA. Each HA VPN gateway has two interfaces, each with its own external IP address, and the SLA depends on using both of them with BGP running over every tunnel. HA VPN supports only dynamic routing with Cloud Router and BGP, so a peer device that can't run BGP can't use it. For the Professional Cloud Network Engineer exam, the HA VPN details worth knowing are what the SLA configuration requires, which peer layouts qualify, and how throughput scales when one tunnel isn't enough.

What the 99.99% SLA requires

To qualify for the 99.99% SLA, an HA VPN gateway needs two active tunnels, one from each of its two interfaces, connected to the peer side. A Cloud Router runs BGP over both tunnels, so if one tunnel fails, traffic moves to the surviving tunnel automatically and nobody has to edit routes by hand.

Both interfaces have to carry a tunnel. A gateway with only one tunnel leaves the second interface unused and falls short of the SLA configuration, even if that interface is kept as a standby. Classic VPN offers 99.9% and doesn't offer the 99.99% SLA at all, however many tunnels or gateways it uses.

Building an HA VPN connection

The setup follows a fixed order.

  • Provision the HA VPN gateway, and create an external VPN gateway resource that represents the on-premises device.
  • Create a Cloud Router in the same region and VPC network as the gateway.
  • Create the tunnels toward the external VPN gateway and associate them with the Cloud Router.
  • Configure a BGP session on each tunnel.

If the peer device can't run BGP, HA VPN can't be used, and custom route advertisements aren't an option either, because they travel over the BGP sessions. Classic VPN with static routing is the option that still works in that case. Once an HA VPN connection is running, setting the VPC network to global dynamic routing lets routes learned in one region reach subnets in every region, so the connection can serve workloads beyond the region where its tunnels end.

Supported peer topologies

HA VPN supports several peer layouts. In Professional Cloud Network Engineer scenarios, the choice between them depends mostly on how many peer devices and public IP addresses the other side has.

  • Two peer devices: each HA VPN interface has a tunnel to a different on-premises VPN device, such as two edge devices at one site.
  • One peer device with one public IP address: create an external VPN gateway resource with a single interface, then build two tunnels, one from interface 0 and one from interface 1, both pointed at that same peer address. Each HA VPN interface has its own external IP, so Google Cloud still has two independent paths, and with BGP sessions on both tunnels this setup qualifies for the 99.99% SLA on the Google Cloud side. It uses one gateway and one peer resource, which makes it the simplest option when the site has one firewall that can run several tunnels and BGP.
  • Another HA VPN gateway: when both sides are in Google Cloud, connect one HA VPN gateway directly to another. This is Google's recommended way to connect two VPC networks over IPsec, and the same two-interface rule applies to the SLA.

Unique interface and peer IP pairings

Google Cloud expects each tunnel to use a unique pairing of a Cloud VPN gateway interface IP address and a peer IP address. Once both HA VPN interfaces already have a tunnel to a single peer IP, more tunnels to that same peer IP repeat a pairing that is already in use and add no usable capacity. To add tunnels, you need a new peer IP address, such as a second on-premises VPN appliance with its own public IP. When the same prefixes are advertised with equal priority over the new BGP sessions, Cloud Router installs equal-cost routes and shares traffic across all of the tunnels.

Scaling Cloud VPN throughput

Every Cloud VPN tunnel has a fixed throughput ceiling of about 250,000 packets per second, which works out to roughly 1 to 3 Gbps depending on packet size. In an active/passive design, the whole connection can only move what one tunnel can move, so once traffic passes that ceiling, the extra packets are dropped.

The supported way to scale Cloud VPN is to add tunnels and make them all active. When the BGP sessions advertise routes with the same priority on both sides, Cloud Router and the on-premises router treat the tunnels as equal-cost paths and use ECMP to hash flows across them. This works when the traffic is made of many parallel flows and the peer supports ECMP. ECMP also needs every BGP session to be up. If one tunnel's session is down, Cloud Router learns routes through only one tunnel and sends all traffic there. Cloud Router does the spreading, so a load balancer plays no part in distributing traffic across tunnels.

Capacity comes from tunnels, so settings that only change how sessions are identified, how the control plane is split or which regions receive learned routes leave the per-tunnel ceiling where it is. Packet size can't be raised to compensate either, because the Cloud VPN MTU can't go above 1460 bytes, and a larger VPC MTU only leads to fragmentation or drops. Additional active tunnels, with ECMP across them, are what add bandwidth.

Our Professional Cloud Network Engineer practice exams cover HA VPN design and throughput alongside Cloud Router BGP configuration and Classic VPN traffic selectors, with detailed explanations that drill these distinctions.

Get tips and updates from GCP Study Hub

arrow