dtrace.conf(16): DTrace state of the union

Slide 1

Slide 1 text

DTrace dtrace.conf(16) CTO [email protected] Bryan Cantrill @bcantrill

Slide 2

Slide 2 text

dtrace.conf(16) • Quadrennial (!) DTrace unconference started in 2008 • ~100 attendees from: • ~45 companies • ~1 VC Entrepreneur-in-Residence • Twitter: #dtraceconf • Thanks to our sponsors, Joyent — and the FIPP! • Huge thanks to Ryan Wilson, Brittany Berry and Jenny Miller from Joyent

Slide 3

Slide 3 text

dtrace.conf prehistory

Slide 4

Slide 4 text

DTrace since 2012

Slide 5

Slide 5 text

DTrace since 2012 Same haircut

Slide 6

Slide 6 text

DTrace since 2012 Same glasses?

Slide 7

Slide 7 text

DTrace since 2012 WTF?!

Slide 8

Slide 8 text

DTrace since 2012

Slide 9

Slide 9 text

DTrace since 2012: Core • Added new ways of representing aggregated data via “agghist,” “aggpack,” and “aggzoom” options • Added new json() subroutine • Added DTrace userland CTF support • Added the new (on-by-default!) “temporal” option • Added print() support for translated types • Added support for fds[], curpsinfo, sched and proc providers in a zone container

Slide 10

Slide 10 text

DTrace since 2012: Platforms • FreeBSD implemented the pid provider in 9.0 and, as of 9.2, enabled DTrace by default! • NetBSD added DTrace support — including support for ARM! • Linux port of DTrace largely completed by Oracle — but keeping the user-level portion proprietary has limited its impact

Slide 11

Slide 11 text

DTrace in 2016 and beyond • Distributed systems are ubiquitous and tracing in distributed systems has improved tremendously; how can DTrace help? • Instrumenting multi-processes applications via the pid provider is still painful; can we improve? • The rise (resurrection?) of statically compiled languages like Go and Rust presents new opportunity — and new challenges • DTrace and serverless computing? • User-level postmortem tracing? • Anti-roadmap: Dynamic translators?