Enhancing Bandaid load balancing at Dropbox by leveraging real-time backend server load information

This post details the enhancement of Dropbox's Bandaid layer-7 load balancing service by incorporating real-time backend server load information. The approach involves piggybacking server utilization (active requests / max concurrent requests) in HTTP response headers (X-Bandaid-Utilization). This allows Bandaid instances to make more informed load balancing decisions using a 'random N choices' method, selecting servers based on their computed scores derived from this utilization metric. The design addresses challenges of distributed load balancing in cloud environments by leveraging readily available application-layer metrics.