LinuxCon Europe + CloudOpen Europe

For years instruction-level debuggers and profilers have improved in leaps and bounds. Similarly, system-level and network monitoring tools are increasingly ubiquitous. There is, however, a gulf between the two worlds which is widening as more and more processes and applications share system resources. In this talk I will present the concepts of application-level tracing and what kind of information is available. I will run through the techniques for combining run-time data from different levels before moving onto some case studies. In particular I will present how a lack of application-level monitoring can lead to file-system bottlenecks and undetected network issues in distributed Linux HPC IT systems.