Measuring and Summarizing Latencies Using the Trace Event Subsystem

Explore the intricacies of measuring and summarizing latencies using the Trace Event Subsystem in this informative conference talk by Tom Zanussi from Intel. Delve into the recently merged patchset in the RT-devel kernel, designed to replace the RT kernel's latency histogram feature with a new version built on a generic 'inter-event' facility. Learn about real-world examples, the underlying mechanics of the event subsystem, and how these new inter-event capabilities integrate with other tracing facilities. Gain insights into potential future developments and enhancements in this field. Discover various aspects of latency measurement, including derived quantities, handlers and actions, chaining handlers, object latencies, and function event magic. Understand how to generate latency histograms for any two events and explore the applications of a general-purpose latency script.

Intro
Trace Events Background (cont'd)
Latency is a derived quantity
Latency (cont'd)
We have a latency value, now what?
Handlers and Actions (onmax and save)
onmax and save output
Chaining handlers (add snapshot())
onmax and save and snapshot output
What about a latency histogram?
onmatch synth and save/snap output
Object latencies (using delete())
Object latencies, part 2
Function Event Magic
Function Event Latency
Generate latency hist for any 2 events
General purpose latency script output
Questions?