[ 
https://issues.apache.org/jira/browse/CLK-306?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13799429#comment-13799429
 ] 

Gilberto C Andrade commented on CLK-306:
----------------------------------------

I've found the problem, I think:
{code:title=PropertyServicePerformanceTest.java|borderStyle=solid}
        public void test_MVELService() throws Exception {
                PropertyService ps = new MVELPropertyService();

                for (int i = 0; i < 100; i++) {
                        Thread testThread = new Thread(new TestRunner(ps, i == 
49));
                        testThread.start();
                }
                
                Thread.sleep(10000);
        }
        
        public void test_OGNLService() throws Exception {
                readDuration.set(0);
                writeDuration.set(0);
                
                PropertyService ps = new OGNLPropertyService();

                for (int i = 0; i < 50; i++) {
                        Thread testThread = new Thread(new TestRunner(ps, i == 
49));
                        testThread.start();
                }
                
                Thread.sleep(10000);
        }

{code}
The for loop in the MVEL test has 100 as limit while in the OGNL has only 50. 
And this complex code to test the performance has some problem: if  you run it 
twice it will invert the winner.

> Replacing OGNL with MVEL
> ------------------------
>
>                 Key: CLK-306
>                 URL: https://issues.apache.org/jira/browse/CLK-306
>             Project: Click
>          Issue Type: New Feature
>          Components: core
>    Affects Versions: 1.4
>         Environment: Target Click 1.5
>            Reporter: Bob Schellink
>            Assignee: Malcolm Edgar
>             Fix For: 2.4.0-RC1
>
>         Attachments: 
> ASF.LICENSE.NOT.GRANTED--expression-language-performance.rar, 
> ASF.LICENSE.NOT.GRANTED--expression-language-performance.zip, mvel.zip, 
> patch.diff
>
>
> I have been following both OGNL and MVEL (another expression language) 
> development the last couple of months. If we are going to make changes to 
> this here are some things you might find interesting:
> * OGNL 2.7.1 which is included in tapestry 4.1, now sports byte code 
> enhancement. This was added by one of the tapestry authors to make OGNL much 
> faster than its reflection mode. However in my testing 2.7.1 threw exceptions 
> every now and then. So still buggy.
> * MVEL (http://mvel.codehaus.org/) seems like a good replacement for OGNL. It 
> has good docs and is actively developed. They even fixed a bug I logged! Like 
> OGNL, MVEL also runs in two modes, either reflection or byte compiled. Unlike 
> OGNL, MVEL's reflection mode is faster than Click's reflection mode.
> I am no expert on this but according to the article below (from MVEL's 
> author), byte code enhancement have some problems in that generated classes 
> accumulate in java's perm space and will only be removed when their 
> classloader is removed.
> http://artexpressive.blogspot.com/2007/07/mvel-by-numbers-real-story.html
> So MVEL in reflection mode looks like an ideal solution here. The power of 
> OGNL with the performance of Click reflection.
> Table columns can again take advantage of expressions. For example to 
> aggregate a total for the row -> new Column("price * tax");



--
This message was sent by Atlassian JIRA
(v6.1#6144)

Reply via email to