5 Replies Latest reply on Oct 31, 2013 12:48 AM by jaikiran

Why is Jboss web http connect so slow? - Over 250ms for a simple servlet/jsp

joshva Oct 29, 2013 11:22 PM

Hi,

I have noticed that Jboss is insanely slow to render a servlet or a simple JSP (example below).

I have tried jboss 6.1 as well as jboss 6.2 beta and wildfly 8.0.0 beta. The same issue occurs in all using 100% default configuration

I have tried using apache mod_jk ajp as well as pure http direct to Jboss. The same issue occurs

I have tried repeatedly calling so its definitely not the JSP compile issue.

The following example renders in 250ms:

173.1.18.180:8082/images/index.jsp

The following example of an image file renders in 50ms (even though it has more content):

173.1.18.180:8082/images/TestImage.jpg

Sample index.jsp code

<%@page contentType="text/html" pageEncoding="UTF-8"%>

<!DOCTYPE html>

<html>

<head>

</head>

<body>

<h1>Hello World!</h1>

</body>

</html>

Any help would be greatly appreciated! We can't run like this

1. Re: Why is Jboss web http connect so slow? - Over 250ms for a simple servlet/jsp

jaikiran Oct 30, 2013 3:47 AM (in response to joshva)

Is that the first access to the JSP? What are the numbers for subsequent access? And how exactly are you measuring those numbers?
Actions
2. Re: Why is Jboss web http connect so slow? - Over 250ms for a simple servlet/jsp

joshva Oct 30, 2013 7:37 AM (in response to jaikiran)

Hi,

It is for subsequent accesses. I am measuring them using: http://www.webpagetest.org/ you could try a variety of other online tools like that. The results are similar. Please note as mentioned if i render a flat image not through the JSP it is 50ms. Even though the bytes being transferred is higher than the JSP's simple HTML code.
Actions
3. Re: Why is Jboss web http connect so slow? - Over 250ms for a simple servlet/jsp

ctomc Oct 30, 2013 8:03 AM (in response to joshva)

Joshua van Aalst wrote:

I am measuring them using: http://www.webpagetest.org/ you could try a variety of other online tools like that.

So basically you are testing network connectivity from some external site, hosted far far far away from your server. But not actually testing response time that your server/application can give you. So is this intentional or you just don't know any other way to test performance?

try using apache bench http://en.wikipedia.org/wiki/ApacheBench
which you can get on any linux distro by installing ab package (on rhel/fedora you can do yum install ab)

and then test your application from same machine you have it on, or at least from same network.

then try testing with

ab -n 100 -c 10 http://localhost:8080/your-page.jsp

this will issue 10 concurrent connection that will each do 100 requests.
and you will get response times from them.

ab is not perfect tool for high load testing but for basic tests like this it is great.

--
tomaz
1 of 1 people found this helpful
Actions
4. Re: Why is Jboss web http connect so slow? - Over 250ms for a simple servlet/jsp

joshva Oct 30, 2013 12:39 PM (in response to ctomc)

Hi Tomaz,

Normally i'd agree with you but not this time...

I have a 57kb image in the war file that you can access here: http://173.1.18.180:8082/images/TestImage.jpg . Using webpagetest.org the single image shows a rendering time of 50ms.

So i do not think it is a network connectivity issue. As the image is 5x faster than the simple jsp both in the same war on the same app server on the same box on the same network.

I will try the ApacheBench software you recommended.

Josh
Actions
5. Re: Why is Jboss web http connect so slow? - Over 250ms for a simple servlet/jsp

jaikiran Oct 31, 2013 12:48 AM (in response to joshva)

Servlets, JSPs and other (real) server side resources unlike static image files obviously involve more processing. So it can potentially take time relatively more than accessing a static image.

Do you have a benchmark for the dynamic resources against which you are comparing the response times? In other words, why do you think 250ms is a bad performance? It may or may not be bad, but we need to understand the benchmark against which you are basing this.
Actions

Go to original post