LinuxCon + ContainerCon North America 2016

When faced with a sporadic problem, our first reflex is to capture as
much Data as possible. Kernel tracing is a great tool to accomplish
this. However, it generates traces that only speak to kernel experts.

We present a workflow based on the lttng-analyses project to extract
meaningful metrics from kernel traces. This information can then be used
to drill down to the root cause of a performance issue. The problem
classes covered include latency, throughput and usage of I/O,
interrupts, scheduling, system calls, CPU, and memory.

We also cover the integration between lttng-analyses and Trace Compass,
which bridges the gap between metrics analysis and interactive navigation
within the traces.