Microsoft just announced the Gateway Load Balancer. This is a new load balancer sku to go along with the existing types of Basic and Standard, however, this is aimed at sending traffic to transparent network devices for in-line traffic inspection. This article will overview the gateway load balancer and explain how F5 BIG-IP can integrate for transparent traffic inspection. Check out MS's blog too.
Many organizations face challenges with Azure network operations, like:
setting up High Availability (HA) can be a learning curve for admins of 3rd party appliances in Azure
using User Defined Routes (UDR's) to direct traffic to a network appliance can be difficult for some organizations, especially updating at scale
dealing with multiple VNets across multiple regions and routing traffic between them can add operational overhead
requiring Source NAT (SNAT) can be a deal-breaker for app owners, but architecting for flow symmetry without SNAT can require serious know-how
Finally, there's a very real skills gap between most Network/Security teams and Cloud teams. Often, a SecOps team member (eg, firewall admin) just wants to manage a firewall, without needing to know about Azure networking. Likewise, the CloudOps team want to send traffic through the firewall without having to alert the SecOps team for every new application added.
With Azure's Basic and Standard load balancer types, it is possible to achieve much of this and I've blogged about this before. But things get complicated quickly with advanced scenarios. The Gateway load balancer makes load balancing network appliances much easier.
It's important to remember that this is aimed at network devices that are inspecting but not proxying traffic. I.e., src and dst IP and port should be preserved when traffic traverses BIG-IP in this scenario, so SNAT is not supported. Think of transparent firewalling as a good use case here.
Enter Gateway Load Balancer
The gateway load balancer is aimed at load balancing firewalls, routers or other network appliances. Some high-level benefits:
It's transparent. Source and Destination IP addresses are unchanged when traffic traverses the gateway load balancer, via VXLAN tunnels to backend pool members. Your firewall admin can now set up policies without worrying about Source or Destination NATs.
Flow symmetry. The load balancer will ensure that response traffic hits the same device on the return path as it did on the request path. This means you can run multiple standalone or Active/Active appliances without using SNAT or UDR's to ensure flow symmetry.
UDR's are not required, because traffic reaches the gateway load balancer in one of two ways:
Either a regular, external load balancer is updated to send traffic via the gateway load balancer before forwarding to it's backend pool, or,
If the VM is not behind an Azure Load Balancer, the VM's NIC is updated so that traffic traverses the gateway load balancer before it receives traffic.
Traffic across VNets and Regions is supported. You can target a gateway load balancer from a regular load balancer that is in a different VNet, or a different region. UDR's are not used, and VNet peering is not required for this traffic flow.
One last important point I'll make: this load balancer type is aimed at solving for North-South traffic, not East-West. I.e., your typical use-case will be for inbound traffic from Internet to your app servers on your Azure VNet. Put your app servers behind a public-facing load balancer, update that load balancer to reference the gateway load balancer, and you're good to go! No need to ask the firewall or F5 admin to configure a new application for you, or a new public IP on their device, etc.
In this scenario we have a web app that is internet-facing. We want to protect the web app from attacks by having the traffic traverse BIG-IP Advanced WAF (AWAF). The gateway load balancer is configured to allow all traffic inbound to the BIG-IP VE's, where security policies are applied. So we deploy a public-facing Azure Load Balancer, configure the load balancer with an attribute referencing the gateway load balancer, and watch as traffic from internet clients is routed via the Advanced WAF devices in transit to the app servers.
A basic demo of this is hosted on GitHub, if you would like to deploy this yourself or see more details. The BIG-IP's will require VXLAN tunnels configuration, VLAN groups, FDB entries, and ARP records. This is to integrate with the VXLAN requirements of the gateway load balancer. See the demo on GitHub here.