Serverless computing is becoming increasingly popular since it reduces management overhead and may reduce costs through pay-per-actual-usage billing. However, to meet the needs of developers, serverless platforms must not only provide low average response time but also low tail latency. To more effectively evaluate serverless clouds, Ustiugov et al. propose STeLLAR, an OSS benchmarking framework to evaluate serverless systems’ performance (end-to-end and per component), including through tail latency-analysis. With STeLLAR, developers can model different load scenarios and serverless application types. The authors’ analysis of the three leading cloud providers (AWS Lambda, Google Cloud Functions, and Azure Functions) finds that storage accesses and bursty function invocations are driving latency variability, while the choice of language runtime does not.