Zscaler design guidance for GRE connectivity emphasizes three key principles: terminate GRE on border (edge) devices, avoid NAT on GRE source addresses, and scale bandwidth by using multiple tunnels. In Zscaler documentation and engineering training, each GRE tunnel is typically sized for up to about 1 Gbps of throughput. For a 2 Gbps requirement, customers are advised to deploy at least two primary GRE tunnels, with two additional backup tunnels for redundancy and failover.

These tunnels should terminate on border routers that own public IP addresses, ensuring optimal routing and simplifying troubleshooting. Zscaler specifically recommends that the public source IPs used for GRE must not be translated by NAT, because the Zscaler cloud must see the original, registered public IP to associate tunnels with the correct organization and enforce policy. Enabling NAT on GRE traffic can break tunnel establishment and lead to asymmetric or unpredictable routing.

Using internal routers introduces extra hops and complexity and often requires NAT or policy-based routing, which goes against recommended best practices. Similarly, any architecture with NAT enabled on GRE traffic conflicts with Zscaler's published requirements. Therefore, the ideal and recommended design for 2 Gbps via GRE is two primary and two backup GRE tunnels from border routers with NAT disabled.