Either an I/O or a range performance issue
------------------------------------------
Key: JRUBY-3001
URL: http://jira.codehaus.org/browse/JRUBY-3001
Project: JRuby
Issue Type: Bug
Components: Performance
Environment: Jruby 1.1.1 to 1.1.4, any OS (Linux, MacOS/X, Windows
XP), and JVM (1.5/1.6), multiple memory (1G to 4G RAM) and cpu configs (1 to 4
processors, Intel/AMD/PPC) and a multitude or JVM settings (-client/-server,
+Xms/+Xmx variations, ParallelGC on/off, etc.)
Reporter: Mike Hoskins
Priority: Minor
Thanks for a wonderful Ruby implementation. How do you get so much done in so
little time? You guys are amazing!
The fact that performance issues (compared to MRI Ruby) are called bugs is
amazing!
Unfortunately, these source code snippets are all slower on jruby (and with
jruby-ng) -
Any size of range (slower whether small or large):
puts (1..10000).to_a
puts (1..100000).to_a
puts (1..1000000).to_a
Or iterated versions of above:
(1..10000).each { |i| puts i }
(1..100000).each { |i| puts i }
(1..1000000).each { |i| puts i }
Even if I remove the puts, it's still far slower (these are about 5-10 times
slower in Jruby 1.1.4 using JDK 1.6.than MRI Ruby 1.8.6):
(1..1000000).each { } # Jruby 1.1.4 - 3.3 seconds; MRI Ruby 1.8.6 - 0.3
seconds
(1..1000000).to_a # Jruby 1.1.4 - 5.9 seconds; MRI Ruby 1.8.6 - 1.0
seconds
(1..10000000).each { } # Jruby 1.1.4 - 7.9 seconds; MRI Ruby 1.8.6 - 1.2
seconds
(1..10000000).to_a # this one seems to cause the JVM to eat memory
(won't fit in 256M JVM), but the MRI Ruby version only takes a 8.6 seconds to
run (Jruby didn't finish in 7 minutes due to memory issues)
I don't know if it's the range or something in the iterators, something in the
JVM, etc.
Now, for the microbenchmarks with numbers (above), I do know that part of it is
the Jruby 1.1.4 startup time. However, this is consistent with Jruby-ng
(1.1.4) and without NailGun. (NailGun brings the smaller microbenchmarks more
in line with MRI, but AFAIK, it's only due to startup time; the longer runs
aren't improved much).
It does this with puts or without, it does it in Windows XP, MacOS/X (Leopard),
and Linux, with JVM 1.5 and 1.6, across Jruby 1.1.1 to 1.1.4, on 1 or more
processors, puts with or without output being redirected to a file, in -client
or -server mode, with or without Parallel GC, etc. It even is slower in
multiple consecutive runs in irb and jirb, where the startup time is minimized.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://jira.codehaus.org/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe from this list, please visit:
http://xircles.codehaus.org/manage_email
