DTrace: Dynamic Tracing in Oracle

Foreword

In early 2004, DTrace remained nascent; while Mike Shapiro, Adam Leventhal, and I had completed our initial implementation in late 2003, it still had substantial gaps (for example, we had not yet completed user-level instrumentation on x86), many missing providers, and many features yet to be discovered. In part because we were still finishing it, we had only just started to publicly describe what we had done—and DTrace remained almost entirely unknown outside of Sun. Around this time, I stumbled on an obscure little Solaris-based tool called psio that used the operating system’s awkward pre-DTrace instrumentation facility, TNF, to determine the top I/O-inducing processes. It must be noted that TNF—which arcanely stands for Trace Normal ...