How to Choose a Google Cloud Load Balancer for the Professional Cloud Network Engineer Exam

GCP Study Hub
November 21, 2026

Google Cloud offers several load balancers, and they differ along four lines: the protocol they understand, whether they proxy or pass traffic through, whether they are external or internal, and whether they are global or regional. Choosing the right one means reading the requirements against those four lines. In most Professional Cloud Network Engineer scenarios, one or two details settle the choice, such as UDP traffic, a need to see the client's own IP address, or a single address that must serve users in several regions.

The four lines that separate the load balancers

Protocol. Application Load Balancers only understand HTTP and HTTPS. Network Load Balancers carry TCP, and some of them also carry UDP. A proprietary TCP protocol, syslog datagrams or RTP audio therefore rule out every Application Load Balancer before any other detail is considered.

Proxy or passthrough. A proxy load balancer ends the client connection and opens a new one to the backend, so the backend sees a proxy address as the source. For an external Application Load Balancer, that connection ends at a Google Front End. A passthrough load balancer forwards packets unchanged and does not end the TCP connection, so the backend sees the client's own source IP address in the packet header.

External or internal. An external load balancer takes traffic from the internet. An internal load balancer gets an internal address from a subnet, so only clients inside the VPC network, or networks connected to it, can reach it. Devices on the public internet cannot reach an internal load balancer directly.

Global or regional. A global load balancer gives one anycast IP address in front of backends in several regions. A regional load balancer serves backends in its own region. Running a regional load balancer in each of several regions produces separate regional addresses, and DNS then has to pick a region for each client, which is not the same as one global address.

The common picks

Combining those four lines gives a short list of load balancers that cover most designs:

  • Global external Application Load Balancer: public HTTP and HTTPS behind one anycast IP address, with backends in several regions, URL-based routing by host and path, Cloud CDN and Google Cloud Armor. It can provide one anycast IPv4 address and one anycast IPv6 address for a global service.
  • Internal Application Load Balancer: a private HTTP tier inside the VPC network, such as an API tier that sits behind a web tier. It runs health checks against its backend service and stops sending requests to unhealthy instances. Pairing an external load balancer for the web tier with an internal one for the middle tier is the standard pattern for multi-tier applications on Google Cloud.
  • Global external proxy Network Load Balancer: non-HTTP TCP traffic from users around the world, behind one anycast address. It ends the client TCP connection at a Google Front End close to the user, then sends the traffic over Google's network to a backend region chosen by proximity, backend health and available capacity.
  • External passthrough Network Load Balancer: internet traffic that must keep the client IP address in the packet header, or any UDP traffic from the internet, such as syslog or RTP audio. It is regional and handles any TCP port.
  • Internal passthrough Network Load Balancer: private non-HTTP TCP or UDP traffic that must keep the client IP address, on one or more ports. It is also the only load balancer that can be used as a route next hop.

Details that decide close cases

Client IP address. When an application or appliance reads the client address from the IP header of each packet, only a passthrough load balancer works, because proxies replace the source address. Proxy load balancers have other ways to carry the address. An Application Load Balancer adds the original client IP address to the X-Forwarded-For header, and a proxy Network Load Balancer can send it with PROXY protocol as an extra header at the start of the stream. Both approaches only help when the backend software reads that header. If the requirement is the address in the IP header, they do not satisfy it.

UDP. UDP support sits with the passthrough Network Load Balancers. Application Load Balancers handle only HTTP and HTTPS, and the external proxy Network Load Balancer terminates TCP only, in both its TCP proxy and SSL proxy forms. Between the two passthrough options, the external or internal line then decides which one internet clients can reach, since the internal one has only an internal address.

Where TLS is decrypted. The two modes of the external proxy Network Load Balancer treat TLS differently. In TCP proxy mode it forwards the TCP stream without decrypting it, so a TLS session passes through untouched and the private keys stay on the VMs. In SSL proxy mode it decrypts at the edge, which requires uploading a certificate to the load balancer. Passthrough load balancers also leave TLS intact, but they are regional.

Regional internal load balancers and other regions. The internal passthrough Network Load Balancer is regional. Global access lets clients in any region reach its frontend, and Cloud DNS routing policies with health checks can steer clients between deployments in different regions. Internal proxy load balancers, by contrast, present the backend with a source address from a proxy-only subnet.

Backends that don't share a template. An unmanaged instance group groups existing VMs that were not created from a shared template, so each VM keeps its own configuration. A managed instance group, in contrast, builds every VM as a copy of one image. The recommended design for an external passthrough Network Load Balancer uses a regional backend service, and that backend service can point to unmanaged instance groups.

Routing is not load balancing

Equal-priority static routes can spread packets across several next hops, but that is routing, not load balancing. These routes do not use load balancer health checks, so traffic can keep flowing to a failed server. They also do not create a frontend that internet clients can reach, because they only influence how traffic inside the VPC network is routed toward next hops. When a design needs health-checked distribution or a public entry point, it needs a load balancer.

For the Professional Cloud Network Engineer exam, we would work through the four lines in order: protocol first, then whether the client IP address must survive, then external or internal, then global or regional. That sequence usually narrows the choice to one load balancer.

Our Professional Cloud Network Engineer practice exams cover choosing a Google Cloud load balancer alongside proxy and passthrough designs and internal load balancing, with detailed explanations that drill these distinctions.

Get tips and updates from GCP Study Hub

arrow