How production engineers support global events on Facebook

This post details the production engineering strategies for supporting global events on Facebook, focusing on handling traffic spikes from high-bandwidth content like Facebook Live. It outlines three categories of load variance (routine, spontaneous, planned) and elaborates on the planning and architectural considerations for planned events like New Year's Eve. The post describes the architecture of Facebook Live, identifying network, CPU, and storage as key resource types affected by increased broadcasts. It highlights the importance of peak concurrent broadcasts as a critical metric for planning and discusses the challenges of accommodating this metric in real-time. The article also details the estimation of New Year's Eve traffic using historical data and growth projections, and describes scaling efforts including hardware allocation and efficiency improvements like consolidating media segments. Finally, it explains three load testing methods: artificially increasing load on tiers, synthetic load tests of dependencies, and shadow traffic.