作者: Eric Koskinen , John Jannotti
关键词:
摘要: Causal request traces are valuable to developers of large concurrent and distributed applications, yet difficult obtain. Traces show how a is processed, can be analyzed by tools detect performance or correctness errors anomalous behavior.We present BorderPatrol, which obtains precise through suystems built from litany unmodified modules. Traced components include Apache, thttpd, PostgreSQL, TurboGears, BIND notably Zeus, closed-source event-driven web server. BorderPatrol using active observation carefully modifies the event stream observed modules, simplifying observation. Protocol processors leverage knowledge about standard protocols, avoiding application-specific instrumentation.BorderPatrol for black-box systems that cannot traced any other technique. We confirm accuracy BorderPatrol's comparing manual instrumentation, compare developer effort required each kind trace. imposes limited overhead on real (approximately 10-15%) it may enabled disabled in at run-time, making viable option deployment production environments.