A buddy of mine sent me Cary Millsap's excellent new paper on the importance of performance diagnostics via extended SQL trace. Cary prefaces, "The trace file contains timings that enable you to precisely measure how long your code takes to run and why.The detailed information about individual database calls and operating system calls is vital to building scalable applications (as well as troubleshooting them in production)."
Here's the link:
http://method-r.com/downloads/doc_details/72-mastering-performance-with-extended-sql-trace
Some interesting items to note:
Page 5 /* free trace library for developers from Method-R */ - "Instrumentation Library for Oracle (ILO), a free open-‐ source project that Method R Corporation maintains at SourceForge. ILO gives you PL/SQL packages that make it dead simple for an application developer to define where in an application’s code path each business task begins and ends. It sets all the “hooks” that your systems DBAs will need later to trace speciaic business tasks by name."
Page 11 /* good info on some 11.1 trace bugs */ - "The Oracle 11.1 era has not been kind to trace file users. Bug 7522002 causes database call time values to run off a 1,024-‐nanosecond-‐per-‐microsecond clock and OS call time values to run off a 1,000-‐ nanosecond-‐per-‐microsecond clock. This causes the sequence of calls in a trace file to zigzag back and forth in time. This problem is correctible, but then bug 8342329 isn’t.It causes the Oracle kernel to emit the same tim value repeatedly for multiple distinct database calls, which is debilitating. Happily, these bugs in 11.1 have patches and are fixed in version 11.2.0.2.0."
No comments:
Post a Comment