Major Clojure 1.10 performance degredation on JDK11
tmarble at info9.net
Wed Feb 6 16:07:06 GMT 2019
Elana Hashman <elana at hashman.ca> writes:
> Mm, given all that, my guess is that this is an AOT compilation issue. I'll
> have to dig deeper into where that's coming from.
Having had this role of measuring JVM performance I can tell you
that this type of "cold start" is tricky to measure accurately.
Obviously we have to factor out machine speed. In the automated
tests we did at Sun we also unmounted the filesystem with the JVM
artifacts to ensure that the file system cache was cold.
It may be that Openjdk 11 has made some startup time tradeoffs
(e.g. more eagerness in favor of better HotSpot behavior).
And Clojure itself may have to initialize a bunch of infrastructure
before evaluating "hi".
Everybody cares about cold start performance, but it will require
more analysis. Getting a set of Clojure tools that work (at any speed)
is a priority for Buster.
More information about the Pkg-clojure-maintainers