We have developed a tool, CRISP (named taking letters from critical and span), to pinpoint and quantify underlying services that impact the overall latency of a request in a microservice architecture. CRISP identifies the critical path in a complex service dependency graph and does so on a large number of traces.
Some really interesting work from Uber. I continue to believe that distributed tracing is a necessary primitive for understanding the behavior of microservice architectures and love to see new ways to use the data to generate insights.