Overview

Distributed load balancing refers to having Envoy itself determine how load should be distributed to the endpoints based on knowing the location of the upstream hosts.

• Load balancing algorithms: Envoy can use several different algorithms to use the provided weights to determine which host to select.

Global load balancing refers to having a single, global authority that decides how load should be distributed between hosts. For Envoy, this would be done by the control plane, which is able to adjust the load applied to individual endpoints by specifying various parameters, such as priority, locality weight, endpoint weight and endpoint health.

A more complicated setup could have resource usage being reported to the control plane, allowing it to adjust the weight of endpoints or to account for the current resource usage, attempting to route new requests to idle hosts over busy ones.

Most sophisticated deployments will make use of features from both categories. For instance, global load balancing could be used to define the high level routing priorities and weights, while distributed load balancing could be used to react to changes in the system (e.g. using active health checking). By combining these you can get the best of both worlds: a globally aware authority that can control the flow of traffic on the macro level while still having the individual proxies be able to react to changes on the micro level.