Am 05.03.2016 um 12:34 schrieb Tullio Bettinazzi:
This is not a memory problem because, otherwise, I'll have the same problem on
all client systems.
It's a communication related problem between server and clients but not
strictly a network problem because, otherwise, two clients, connected to the
same cable would perform in the same way.
If it is stable slow on one client, then you could take threaddumps on
the tomcat side and look what it is doing. On the network side, you
could look at a tcpdump.
Have you tried switching the connectors on the tomcat side?
Felix
Tks
Tullio
Subject: Re: Performance regression from 7 to 8
To: users@tomcat.apache.org
From: felix.schumac...@internetallee.de
Date: Sat, 5 Mar 2016 11:13:58 +0100
Am 04.03.2016 um 14:19 schrieb Tullio Bettinazzi:
Done and nothing changed.
Any suggestion ?
It could be related to memory usage.
Tomcat 8 can use more memory than tomcat 7 (See
https://mail-archives.apache.org/mod_mbox/tomcat-users/201602.mbox/%3ccacbju2wmw7mntevb6hwjqdfzsjpmfiuw6k_dn1u0ufh0haj...@mail.gmail.com%3E)
So try to look at your memory consumption and adjust the limits for the
jvm accordingly. For monitoring, you can enable gc logging, or use
something like jstat, jconsole, jvisualvm, jmc or any other monitoring tool.
Mark has worked on the memory issue and lowered consumption for newer
versions. I think they will be in the next release.
Regards,
Felix
Here the code.
package axioma.rubik.engine.web.servlet;
import java.io.*;
import javax.servlet.ServletException;
import javax.servlet.annotation.WebServlet;
import javax.servlet.http.*;
@WebServlet(name="Test8", description="Direct update of data",
urlPatterns={"/Test8"})
public class Test8Servlet extends HttpServlet {
private static final long serialVersionUID = 1L;
@Override
protected void doGet(HttpServletRequest request, HttpServletResponse
response) throws ServletException, IOException {
try {
fai(response);
} catch (Exception ex) {
ex.printStackTrace();
}
}
public void fai(HttpServletResponse response) throws IOException {
ByteArrayOutputStream bbs = new ByteArrayOutputStream();
BufferedOutputStream bos = new BufferedOutputStream(bbs);
for(int i = 0; i < 400000; i++) {
bos.write(96);
}
bos.flush();
bbs.writeTo(response.getOutputStream());
}
}
Date: Fri, 4 Mar 2016 12:58:02 +0100
Subject: Re: Performance regression from 7 to 8
From: r...@apache.org
To: users@tomcat.apache.org
2016-03-04 12:42 GMT+01:00 Mark Thomas <ma...@apache.org>:
On 04/03/2016 11:17, Tullio Bettinazzi wrote:
This servlet reproduces the problem perfectly.
Getting better but still some room for improvement.
- You don't need to implement doPost()
- You don't need to call System.gc() (or if you do look there for
the problem)
Yes, it's on every get and will cause a major concurrency issue.
- You do need to remove the use of the ComunicationChannelHttp and
Cronometro classes (and if that fixes the problem look there
for the root cause)
- The try/catch in doGet() should not be necessary either
Also writing individual bytes is more costly even if there's some buffering.
Rémy
Mark
package axioma.rubik.engine.web.servlet;
import java.io.*;
import javax.servlet.ServletException;
import javax.servlet.annotation.WebServlet;
import javax.servlet.http.*;
import axioma.rubik.engine.web.ComunicationChannelHttp;
import it.axioma.rubik.engine.Cronometro;
@WebServlet(name="Test8", description="Direct update of data",
urlPatterns={"/Test8"})
public class Test8Servlet extends HttpServlet {
private static final long serialVersionUID = 1L;
@Override
protected void doPost(HttpServletRequest request,
HttpServletResponse response) throws ServletException, IOException {
this.doGet(request,response);
}
@Override
protected void doGet(HttpServletRequest request, HttpServletResponse
response) throws ServletException, IOException {
try {
fai(response);
System.gc();
} catch (Exception ex) {
ex.printStackTrace();
}
ComunicationChannelHttp.CONTEXT_MANAGER.clean();
}
public void fai(HttpServletResponse response) {
Cronometro crono = new Cronometro();
ByteArrayOutputStream bbs = new ByteArrayOutputStream();
BufferedOutputStream bos = new BufferedOutputStream(bbs);
try {
for(int i = 0; i < 400000; i++) {
bos.write(96);
}
bos.flush();
System.out.println("Step 1 : "+crono.elapsed());
bbs.writeTo(response.getOutputStream());
System.out.println("Step 1 : "+crono.elapsed());
} catch (IOException ex) {
ex.printStackTrace();
}
}
}
Subject: Re: Performance regression from 7 to 8
To: users@tomcat.apache.org
From: ma...@apache.org
Date: Fri, 4 Mar 2016 10:38:30 +0000
On 04/03/2016 10:24, Tullio Bettinazzi wrote:
The problem is all in this small piece of code
ByteArrayOutputStream bbs = new ByteArrayOutputStream();
BufferedOutputStream bos = new BufferedOutputStream(bbs);
trans.eseguiTrasformazioneOut(bos);
try {
bos.flush();
initReponse(xpFileTypeOut.getMimeType(), xpFilename);
bbs.writeTo(getOutputStream());
} catch (IOException ex) {
Messaggi.getErrori().getLogger().error("Errore in
emettiFile ", ex);
}
The yellow instruction take 100 ms in Tomcat7, quite stable on all
clients, in Tomcat8 it takes from 50 ms to 4500 ms stable on a single
client PC but very different from client to client.
Tks
Tullio
I'll repeat what I said previously:
Try creating the *simplest possible* web application that demonstrates
the
problem.
Mark
Subject: Re: Performance regression from 7 to 8
To: users@tomcat.apache.org
From: ma...@apache.org
Date: Fri, 4 Mar 2016 09:42:22 +0000
On 04/03/2016 09:39, Tullio Bettinazzi wrote:
I applied tour suggestion and analyzed, with firebug, the
composition of the time.
The difference between 7 and 8 is the "receiving" time which is more
or less zero in 7 and 2sec. in 8.
How can I understand such difference ?
Try creating the simplest possible web application that demonstrates
the
problem.
Mark
Tks
Tullio
P.S. : same server, same client, same network, same code both 7 and
8 installed from scratch
Subject: Re: Performance regression from 7 to 8
To: users@tomcat.apache.org
From: geor...@mhsoftware.com
Date: Thu, 3 Mar 2016 09:30:33 -0700
On 3/3/2016 4:06 AM, Tullio Bettinazzi wrote:
I've an application in which I write a page from a Buffered Stream
directly to the Servlet output stream (more or less 300kb).
In 7 it works perfectly (100ms).
In 8 , depending from the network connection and mainly from the
http client itself (the browser in the PC) the same operation
takes from
50ms to 4500 ms.
One of the things I would look at is the browser debug window. Open
the
debugger, and go to the Networks/Timings tab and load both pages.
That
would give some insights as to what's happening. Perhaps it is the
page.
Perhaps there's something else.
On the same PC I find more or less the same time using Chrome and
Firefox also changing network connections (wifi, lan, adsl).
Could someone suggest a solution ?
Tks
Tullio
--
George Sexton
*MH Software, Inc.*
Voice: 303 438 9585
http://www.connectdaily.com
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org
For additional commands, e-mail: users-h...@tomcat.apache.org
