Where should distributed load balancing occur in a horizontally scalable architecture?

Firewall-side load balancing is primarily a security function and not typically where advanced distributed load balancing for horizontally scalable applications occurs.

Network-side/central load balancing uses a single point or cluster of load balancers, which can become a bottleneck in highly distributed and horizontally scaled architectures.

While service-side load balancing (e.g., within a service mesh) is a distributed approach, client-side load balancing typically refers to the initiator of the request making the balancing decision, which is a more direct form of distributed load balancing at the request origin.

Client-side/local load balancing distributes the load balancing decision-making to each individual client, making the load balancing mechanism itself horizontally scalable and avoiding a central bottleneck. This model often relies on service discovery and allows clients to directly connect to chosen backend instances, which aligns with distributed, horizontally scalable architectures.