Dapper (paper) conceptpaper
Sigelman et al.
Sigelman et al. (Google), 2010. The paper that established the model for distributed tracing — trace context propagated through every RPC, sampled for performance, reconstructed into a tree of spans. OpenTelemetry, Jaeger, Zipkin, and every commercial APM are descendants. Two key insights: (1) tracing has to be ubiquitous and low-overhead to be useful, which means sampling; (2) sampling needs to be consistent within a trace or you get useless partial pictures.