JBoss EL performance vs Sun EL| JBoss.org Content Archive (Read Only)

15. Re: JBoss EL performance vs Sun EL

christian.bauer Sep 17, 2007 3:39 PM (in response to mgrouch)

Why is everybody guessing what and where a performance problem might be? It would be very easy indeed to just create a snapshot with a profiler and posting it here.

16. Re: JBoss EL performance vs Sun EL

mgrouch Sep 18, 2007 12:23 AM (in response to mgrouch)

To avoid guess here is a test case

import java.io.Serializable;

import org.jboss.seam.annotations.Name;

@Name("testBean")
public class TestBean implements Serializable {

 public final static int ROWS = 1000;
 public final static int COLS = 10;
 public final static int ITEMS = 100;

 private String data[][][] = new String[ROWS][COLS][ITEMS];

 public TestBean() {
 for (int i=0; i<ROWS; i++) {
 for (int j=0; j<COLS; j++) {
 for (int k=0; k<ITEMS; k++) {
 data[j][k] = "[" + i + "," + j + "," + k + "]";
 }
 }
 }
 }

 public String[][][] getData() {
 return data;
 }

 public void setData(String[][][] data) {
 this.data = data;
 }
 }

JSF

 <table>
 <tbody>
 <tr>
 <ui:repeat value="#{testBean.data}" var="row" varStatus="index">
 <td>
 <ui:repeat value="#{row}" var="col">
 <select>
 <ui:repeat value="#{col}" var="item">
 <option value="#{item}">#{item}</option>
 </ui:repeat>
 </select>
 </ui:repeat>
 </td>
 </ui:repeat>
 </tr>
 </tbody>
 </table>

JSP

<%@page import="com.mydomain.biz.bean.TestBean"%>
<html>
<head>
</head>
<body>
<table>
<%
 TestBean bean = new TestBean();
 for (int i=0; i<TestBean.ROWS; i++) {
%>
 <tr>
<%
 for (int j=0; j<TestBean.COLS; j++) {
%>
 <td>
 <select>
<%
 for (int k=0; k<TestBean.ITEMS; k++) {
 String data = bean.getData()[j][k];
 %>
 <option value="<%=i%>_<%=j%>_<%=k%>"><%=data%></option>
 <%
 }
 %>
 </select>
 </td>
 <%
 }
 %>
 </tr>
 <%
 }
 %>
 </table>
 </body>
 </html>

I don't have profiler handy, but I swear it was EL functions which topped the list in it.

17. Re: JBoss EL performance vs Sun EL

halivingston Sep 19, 2007 10:04 AM (in response to mgrouch)

Are we addressing these issues in JSF 2.0? If the culprit is JSF i.e.

Disappointing if I have to use Wicket, because Seam is just looking so good for my project. But the demos on my 2.6Ghz machine run slow, as compared to say a PHP equivalent of those applications (seam-booking)

So, will Seam outperform PHP, and the likes under stress? Because single execution the PHP pages outperform Seam by a stretch?

18. Re: JBoss EL performance vs Sun EL

gavin.king Sep 19, 2007 6:31 PM (in response to mgrouch)

1) In JSF for N mutually exclusive 'rendered' conditions they will ALWAYS be evaluated N times!!!

Well, sure.

But note that a switch statement is also involves n potential evaluations. I'm not sure what the difference is? I guess you can emulate a switch using nested f:subviews.

If this particular issue is truly a problem, it is possible to decide which fragment to render in your Java code by using rendered="false" on all N fragments and calling UIComponent.setRendered(true) explicitly before the render phase.

2) Even worse there is no syntax to assign columnBean.type to a variable so it will be EVALUATED EACH TIME!!!

In components.xml, type:

<factory name="columnBeanType" scope="request" value="#{columnBean.type}"/>

Perhaps that "trick" addresses your concerns?

3) Even worse it will be evaluated using REFLECTION!!!

It has been a really long time since a reflective method invocation was significantly slower than a normal method call. This seems to be a red herring.

4) JSF will give you temptation to write condition as STRING comparison

I don't know what you mean.

19. Re: JBoss EL performance vs Sun EL

gavin.king Sep 19, 2007 6:35 PM (in response to mgrouch)

"halivingston" wrote:
So, will Seam outperform PHP, and the likes under stress? Because single execution the PHP pages outperform Seam by a stretch?

Seam is definitely not going to outperform something like PHP for simple page rendering. The level of abstraction is so much higher!

However, there are two reasons why Seam can beat PHP in more advanced cases:

* Seam can do much more efficient caching of state, avoiding hits to the database
* The Java EE server is just, by nature, more scalable

20. Re: JBoss EL performance vs Sun EL

mgrouch Sep 19, 2007 6:55 PM (in response to mgrouch)

Gavin,

My critique was addressed to JSF not Seam.
I think Web Remoting in Seam might look more promising than JSF.
JSF IS slow. Java component development in it requires lots of labor.

What do you think about web remoting with something like

http://extjs.com/

which open source LGPL licensed.
Demo is here http://extjs.com/.

What would think is more appealing?

Thanks

21. Re: JBoss EL performance vs Sun EL

mgrouch Sep 19, 2007 10:51 PM (in response to mgrouch)

Here are some results of performance testing:

With the code I've posted in this thread (I've modified it to put
100 rows, 100 columns, 100 items in dropdowns).

Rendering table using JSP

$ time curl http://localhost:8080/TestJSP/testTable.jsp > foo.jsp
 % Total % Received % Xferd Average Speed Time Curr.
 Dload Upload Total Current Left Speed
100 49.4M 0 49.4M 0 0 6025k 0 --:--:-- 0:00:08 --:--:-- 3884k

real 0m9.566s
user 0m0.124s
sys 0m0.765s

Rendering table using Seam/JSF

$ time curl http://localhost:8080/Table/testTable.seam > foo.seam
 % Total % Received % Xferd Average Speed Time Curr.
 Dload Upload Total Current Left Speed
100 55.1M 0 55.1M 0 0 1055k 0 --:--:-- 0:00:53 --:--:-- 756k

real 0m53.694s
user 0m0.343s
sys 0m0.671s

JSP rendering took 9.5 seconds
JSF rendering took 53 seconds

I've made many runs with JSP consistently outperforming JSF by 5 times.

Tests were done on HP-A1130N PC.
I was monitoring memory usage (no excessive swapping)

JSF application is seam-gen generated (server side state saving,
standard configuration), JBoss from RedHat IDE.

These results are consistent with what I'm observing in our QA environment.

For similar table pages which would render in 1 sec we are getting responses in around 5 sec.

So performance of rendering tables is not sufficient.
Regards

22. Re: JBoss EL performance vs Sun EL

dustismo Sep 20, 2007 1:07 PM (in response to mgrouch)

We had similar issues with rendering large tables. The problem was fixed by simply outjecting the variables instead of accessing them through the bean.

So in you example try:

@Out
public String[][][] getData() {
 return data;
}

and in the view:

<ui:repeat value="#{data}" var="row" varStatus="index">

hope that helps,
Dustin

23. Re: JBoss EL performance vs Sun EL

mgrouch Sep 20, 2007 2:14 PM (in response to mgrouch)

Thanks, we will try that.

However I think JBoss EL implementation has to be performance tuned anyway.
For a table I still would have to write

#{entity.property1}
#{entity.property2}
#{entity.property3}
#{entity.property4}

...

and so on for each column.
And outjection doesn't help with these EL expressions.

Thanks

24. Re: JBoss EL performance vs Sun EL

mgrouch Sep 20, 2007 2:26 PM (in response to mgrouch)

My point with these code samples wasn't to write efficient code.
Actually it was quite opposite - write ineffifient code to create stress conditions so that ineffifiencies on internal implementation
EL/JSF/Seam would show up more clearly in profiler.
And that would help in fixing it.

Thanks

25. Re: JBoss EL performance vs Sun EL

mgrouch Sep 21, 2007 12:09 AM (in response to mgrouch)

"dustismo" wrote:
We had similar issues with rendering large tables. The problem was fixed by simply outjecting the variables instead of accessing them through the bean.

So in you example try:
@Out
public String[][][] getData() {
 return data;
}
and in the view:
<ui:repeat value="#{data}" var="row" varStatus="index">
hope that helps,
Dustin

I've tried that. It didn't help.
Same slow response times.

26. Re: JBoss EL performance vs Sun EL

mgrouch Sep 23, 2007 1:40 PM (in response to mgrouch)

Can you use cglib's fast method invocation in JBoss EL?

Thanks

27. Re: JBoss EL performance vs Sun EL

christian.bauer Sep 23, 2007 1:58 PM (in response to mgrouch)

Is there a particular reason why you refuse to use a profiler?

28. Re: JBoss EL performance vs Sun EL

mgrouch Sep 23, 2007 3:21 PM (in response to mgrouch)

Yes. I don't have it right now.
BTW:
I've tried cglib FastClass and it is faster than what you are doing in JBoss EL
ReflectionUtil.
Thanks

29. Re: JBoss EL performance vs Sun EL

fhh Sep 23, 2007 4:47 PM (in response to mgrouch)

Well, if you don't have a profiler you should at least test EL performance only. Instead of building a tree of 1.000.000 components you could simply loop 1.000.000 times over a single EL evaluation (no ui component). That would exclude JSF from the list of suspects.

Regards

Felix