Simple load balancing methods include random and round-robin, where LB instances determine the number of requests sent to each server solely based on the properties of the scheduling algorithms. At the end, we present several high-level approaches to tackle these challenges. In this section, we first review the traditional load balancing methods and then discuss the challenges these methods face in a cloud environment where load balancing is performed distributedly. Load balancing in the networking area has been well studied and many load balancing methods exist. By sharing our experiences and results obtained from the Dropbox production environment, we hope to contribute an additional data point showing the effectiveness of these approaches and highlight new challenges in load balancing. When we first read their post while working on this project, it was encouraging to see a similar approach being explored, and an interesting technical problem that attracts engineers in this field. It’s worth noting that the main idea we employed in this work has been presented in Netflix’s article Rethinking Netflix’s Edge Load Balancing. In this post, we will share the experiences and results, as well as discussing load balancing in general. Our Traffic/Runtime team recently explored leveraging real-time load information from backend servers to make better load balancing decisions in Bandaid. More balanced load distribution among backend servers is desirable because it improves the performance and reliability of our services which benefits our users.

In an earlier post, we introduced Bandaid, the service proxy we built in house at Dropbox that load balances the vast majority of our user requests to backend services. Layer-7 load balancing (LB) is a core building block to scale today’s web services.

Dropbox cost details server labor