Top Ten Metrics for Evaluating your Garbage Collector

[Christine Flood]

Title:  Top Ten Metrics for Evaluating your Garbage Collector

Abstract:

There's an old joke about a new bride who prepares her first big
dinner for her new husband.  She's making her mother's brisket recipe
which starts by cutting the ends off of the roast.  Her husband asks
her why she cuts the ends off. She says she does it because that's the
way her mother always did it.  She asks her mother and her mother says
"That's the way your grandmother always did it".  She asks her
grandmother and her grandmother says "That's the only way it would fit
in my pan".  There's a lot of folklore and myth surrounding Java
garbage collection performance, most of which is obsolete, irrelevant
to your application, or was never really right in the first place.

So how do we evaluate a GC algorithm?  We can measure various metrics:
 end to end run time, maximum pause time, average pause time,
percentage of the cpu given to the Java threads, memory footprint,
micro benchmarks, ...  This talk will focus on ten (give or take)
metrics/tools we can use to evaluate garbage collection performance.
This isn't a comparison of OpenJDK collectors, it's a discussion of
what characteristics are important in a good garbage collector and how
to measure them.


Speaker:  Christine H. Flood is one of the authors of Shenandoah at
Red Hat.  She has been hacking Garbage Collectors for longer than she
cares to admit.  She's published academic papers on Shenandoah, G1,
and the Parallel Collector and is a Java One Rock star.

Time:  25 minutes

Recording video and/or audio CC-BY-2.0 license is acceptable.