JSON serialisation

raubarede · March 17, 2010, 11:26pm

Hello,

I need to make a json serialisation which should work with any data
coming from my jruby (embedded) code.

I use SOJO framework which is ok.
But JSON has some constraints :

Object keys must be String
Date Time cant be coded, whe need convert them in String
…

So i realise this little function which prepare the data to be
serialized
(code in string any data which JSON dont know) :

def to_serialise(o)
case o
when Java::JavaUtil::ArrayList then o.map { |no| to_serialiseno)
}
when Java::JavaUtil::HashMap then o.inject({}) {
|h,kv| h[kv[0].to_s] = to_serialisekv[1]) ; h
}
when Array then o.map { |no| to_serialise(no) }
when Hash then o.inject({}) {
|h,kv| h[kv[0].to_s] = to_serialise(kv[1]) ; h
}
when Fixnum then o
else
o.to_s
end
end

All work find, but CPU time is realy excessive (15 ms traitment become
180 ms with
little Hash …)
So i wonder howto write that in Java : i don’t known how iterate on
Hash/Array org.jruby
in java…

Can someone help me ?

Thanks,
RÃ©gis

raubarede · March 18, 2010, 3:28am

I think your paste got a little messed up, so I made a cleaned-up
version plus a benchmark here: serializer.rb · GitHub

For this benchmark, it takes a bit less than 100ms per iteration for a
10000-element hash containing string keys and hash values which
themselves have fixnum keys and array values containing fixnums. So
the structure is roughly:

10000 * {string => {fixnum => [fixnum]]

That performance seems reasonable, and that seems to be a reasonably
large hash. Here’s the benchmark numbers on my system with Java 6 in
client mode:

~/projects/jruby âž” jruby serializer.rb
0.447000 0.000000 0.447000 ( 0.422000)
0.100000 0.000000 0.100000 ( 0.100000)
0.100000 0.000000 0.100000 ( 0.100000)
0.135000 0.000000 0.135000 ( 0.135000)
0.099000 0.000000 0.099000 ( 0.099000)

In server mode, it’s better still:

~/projects/jruby âž” jruby --server serializer.rb
1.046000 0.000000 1.046000 ( 0.979000)
0.357000 0.000000 0.357000 ( 0.357000)
0.060000 0.000000 0.060000 ( 0.060000)
0.062000 0.000000 0.062000 ( 0.062000)
0.056000 0.000000 0.056000 ( 0.056000)

Granted, I don’t have any code hitting ArrayList or HashMap instances.
Perhaps that’s where your performance bottleneck comes from? There are
improvements in JRuby 1.4 and again in JRuby 1.5 for calling out to
Java code. For example, the cost of iterating a 10000-element
ArrayList in the past three JRuby versions:

gist.github.com

https://gist.github.com/headius/335979

gistfile1.txt

~/projects/jruby ➔ ../jruby-1.3.1/bin/jruby --server -rjava -rbenchmark -e "a = java.util.ArrayList.new; 100000.times { a << java.lang.Object.new }; 5.times { puts Benchmark.measure { a.each { |x| x } } }"
  0.768000   0.000000   0.768000 (  0.716000)
  0.328000   0.000000   0.328000 (  0.328000)
  0.240000   0.000000   0.240000 (  0.241000)
  0.239000   0.000000   0.239000 (  0.239000)
  0.254000   0.000000   0.254000 (  0.254000)

~/projects/jruby ➔ ../jruby-1.4.0/bin/jruby --server -rjava -rbenchmark -e "a = java.util.ArrayList.new; 100000.times { a << java.lang.Object.new }; 5.times { puts Benchmark.measure { a.each { |x| x } } }"
  0.803000   0.000000   0.803000 (  0.748000)
  0.226000   0.000000   0.226000 (  0.226000)

This file has been truncated. show original

So it’s also possible your performance issues may improve with newer
versions anyway.

If you really needed to write this in Java, it’s not going to be too
difficult. In JRuby, a Ruby Hash implements java.util.Map and a Ruby
Array implements java.util.List, so code that can walk those types
will work fine. But I’d ideally like to make it possible for you to
leave the code in Ruby, so we should work together to see if we can
speed it up.

I’d recommend giving JRuby 1.5 dev snapshots a try:
http://ci.jruby.org/snapshots

On Wed, Mar 17, 2010 at 4:26 PM, Regis d’Aubarede [email protected]
wrote:

Â Â Â Â }
All work find, but CPU time is realy excessive (15 ms traitment become

Posted via http://www.ruby-forum.com/.

To unsubscribe from this list, please visit:

Â Â http://xircles.codehaus.org/manage_email

To unsubscribe from this list, please visit:

http://xircles.codehaus.org/manage_email

raubarede · March 18, 2010, 6:43pm

Hello,

Thank you for your benchmark.
I have rewrite the function in java :

gist.github.com

https://gist.github.com/glurp/336594

to_jsonable in java

	static public Object to_jsonnable(Object o) {
	  if (o instanceof List) { 
		List lo = (List)o;
		ArrayList nlo = new ArrayList();
		for (Object a : lo) nlo.add(to_jsonnable(a)) ;
		return(nlo);
	  } else if (o instanceof Map) {
			Map mo = (Map)o;
			LinkedHashMap<Object, Object> nmo =
                            new LinkedHashMap<Object, Object>();

This file has been truncated. show original

Effectively, performances differences are not so big
(sorry, i am embedded in a server container, i have not
try Benhmark class…):

trace with ruby function :

traces with java function

I have embeded my JRuby/JSON server in a Jetty server, which is embedded
in a scada system.
and now i work in respond/generate data for a ExtJs Crud application in
navigator…

by

raubarede · March 18, 2010, 7:20pm

On Thu, Mar 18, 2010 at 12:43 PM, Regis d’Aubarede
[email protected] wrote:

Hello,

Thank you for your benchmark.
I have rewrite the function in java :

to_jsonable in java · GitHub

Effectively, performances differences are not so big
(sorry, i am embedded in a server container, i have not
Â try Benhmark class…):

Yes, the performance seem to be about the same. To me this means that
either the Ruby collections are slowing things down equally in both
cases, or else there’s just a lot of data to walk and the walking
itself isn’t slowing things down. You probably should try running a
Java profiling tool if this performance is not good enough…it may
help you find where the problem lies. If it’s in JRuby’s collections,
we can see about improving them.

Charlie

To unsubscribe from this list, please visit:

http://xircles.codehaus.org/manage_email

JSON serialisation

Â Â Â Â } All work find, but CPU time is realy excessive (15 ms traitment become

trace with ruby function :

traces with java function

Â Â Â Â }
All work find, but CPU time is realy excessive (15 ms traitment become