You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@mina.apache.org by Steve Johns <st...@gmail.com> on 2008/04/02 02:44:20 UTC
Write slow or OOM?
So far Mina does a good job in my recent project until yesterday. I am
working a GW server handling 10k - 15k incoming messages/sec with NOT very
big sizes and send them to clients. The weird thing is the program has NO
problem to consume the incoming messages(decoding, read). However, when a
simple telnet client connected to my server(same machine), I observed the GC
taking 0.5 - 1.0sec to do GC every time and eventually OOM(GC is NOT
faster??). I got serveral questions:
1) just wonder if mina is able to write that many of messages? I believe
telnet client on the same machine NOT considered to be a SLOW client, right?
2) IF 1) Yes, the problem seems GC limitation but NOT Mina problem?
Thanks.
Re: Write slow or OOM?
Posted by Steve Johns <st...@gmail.com>.
I made some mistakes on my former example. Sorry. Let me post new result
when done. Thanks.
2008/4/2 Steve Johns <st...@gmail.com>:
> Thank you very much.
>
> 1) I used Mina1.1.5 NOT 2.0 M1.
> 2) Actually telnet client is on the same machine as my Mina server. Could
> it be SLOW?
>
> In order to test how fast Mina could write, I modified the
> ReverseProtocolHandler example. I added following code in the
> ReverseProtocolHandler:
>
> public void sessionCreated(IoSession session) throws Exception {
> String message =
> "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
>
> while (true) {
> for (int i = 0; i < 5000; i++)
> session.write(message);
>
> Thread.sleep(500);
> }
> }
>
> and I opened a telnet client connect to it. However, when I
> PrintGCDetails, i obversed that GC is so busy to collect (take 0.15 sec to
> collect once, and keep collecting )and after a while it seems too slow to
> collect. Eventually OOM.
>
> I don't know if my code added was correct or NOT. I was impressed when you
> said your server could handle 140k message/sec. How could you possible do
> that? ^^
> On Wed, Apr 2, 2008 at 11:30 AM, "이희승 (Trustin Lee) <tr...@gmail.com>
> wrote:
>
> > Hi Steve,
> >
> > OOM can be raised when the write request rate is so high. There are two
> > possible issues:
> >
> > 1) MINA NioProcessor is writing slowly.
> > 2) Receiving party (i.e. telnet client) is receiving slowly.
> >
> > If #1 is the case, we need to fix it. However, I was able to achieve
> > 140k messages per second for asynchronous messaging and 20k messages per
> > second for synchronous messaging in my private performance test
> > performed in gigabit network recently. Just in case your scenario is
> > triggering a hidden bug in MINA, please provide us a simple test
> > application.
> >
> > HTH,
> >
> > Steve Johns wrote:
> > > So far Mina does a good job in my recent project until yesterday. I am
> > > working a GW server handling 10k - 15k incoming messages/sec with NOT
> > very
> > > big sizes and send them to clients. The weird thing is the program has
> > NO
> > > problem to consume the incoming messages(decoding, read). However,
> > when a
> > > simple telnet client connected to my server(same machine), I observed
> > the GC
> > > taking 0.5 - 1.0sec to do GC every time and eventually OOM(GC is NOT
> > > faster??). I got serveral questions:
> > > 1) just wonder if mina is able to write that many of messages? I
> > believe
> > > telnet client on the same machine NOT considered to be a SLOW client,
> > right?
> > > 2) IF 1) Yes, the problem seems GC limitation but NOT Mina problem?
> > >
> > > Thanks.
> > >
> >
> > --
> > Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> > --
> > what we call human nature is actually human habit
> > --
> > http://gleamynode.net/
> >
> >
>
Re: Write slow or OOM?
Posted by Steve Johns <st...@gmail.com>.
I found out that windows telnet should be considered a "slow" client which
caused my server OOM. After I used my own decode client, I got
5000messages/sec(10kmessages/sec should be okay). Thanks Mina and Trustin.
2008/4/3 Steve Johns <st...@gmail.com>:
> Thanks a lot.
>
> Actually I fount out Thread.sleep(1) didn't work as what I expected. On
> Windows XP, the minimum time Thread Acutally could sleep is 15ms. Which
> meant my example didn't really simulate messages as much as expected. What I
> tried to do was to generate 5k - 10k messages/sec and send them to the
> client.
>
> On Thu, Apr 3, 2008 at 5:55 PM, "이희승 (Trustin Lee) <tr...@gmail.com>
> wrote:
>
> > I was not able to reproduce OOM in my laptop (Intel Core 2 Duo T9300
> > 2.5GHz 6MB L2 Cache, 2GB RAM, JDK 1.6.0.10-beta). The following is the
> > source code I have used - it was simply ported to MINA 2:
> >
> > import java.net.InetSocketAddress;
> > import java.nio.charset.Charset;
> >
> > import org.apache.mina.common.IoHandlerAdapter;
> > import org.apache.mina.common.IoSession;
> > import org.apache.mina.filter.codec.ProtocolCodecFilter;
> > import org.apache.mina.filter.codec.textline.TextLineCodecFactory;
> > import org.apache.mina.transport.socket.SocketAcceptor;
> > import org.apache.mina.transport.socket.nio.NioSocketAcceptor;
> >
> > public class MyHandler extends IoHandlerAdapter {
> > IoSession mySession;
> >
> > public static void main(String[] args) throws Exception {
> > SocketAcceptor acceptor = new NioSocketAcceptor(2);
> >
> > // Prepare the configuration
> > acceptor.getFilterChain().addLast(
> > "codec",
> > new ProtocolCodecFilter(new TextLineCodecFactory(Charset
> > .forName("UTF-8"))));
> > acceptor.getSessionConfig().setTcpNoDelay(true);
> >
> > // Bind
> > acceptor.setHandler(new MyHandler());
> > acceptor.bind(new InetSocketAddress(8080));
> >
> > System.out.println("Listening on port 8080");
> > }
> >
> > @Override
> > public void exceptionCaught(IoSession session, Throwable cause) {
> > cause.printStackTrace();
> > // Close connection when unexpected exception is caught.
> > session.close();
> > }
> >
> > @Override
> > public void sessionCreated(IoSession session) throws Exception {
> > mySession = session;
> >
> > new MyThread().start();
> > }
> >
> > private class MyThread extends Thread {
> > @Override
> > public void run() {
> > String message =
> > "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
> >
> > while (true) {
> >
> > mySession.write(message);
> > mySession.write(message);
> > mySession.write(message);
> >
> > try {
> > Thread.sleep(1);
> > } catch (InterruptedException e) {
> > // TODO Auto-generated catch block
> > e.printStackTrace();
> > }
> > }
> > }
> > }
> > }
> >
> > On the client side, I used netcat:
> >
> > $ nc localhost 8080
> >
> > Please note my test was run in Linux. I'm not sure about Windows
> > unfortunately. There's also netcat for Windows IIRC - you could try
> > that instead of Windows telnet, which doesn't seem to perform very well.
> >
> > HTH,
> >
> > Steve Johns wrote:
> > > I wrote the following testing program. I used one thread to simulate
> > the
> > > incoming datafeed. Sleep(1) = 1000messages/sec(I added 3
> > > session.write(message)), then 3000 messages / sec roughly sent to
> > client.
> > > My server CPU reaches 70% with one telnet client on the same machine.
> > >
> > -----------------------------------------------------------------------------------
> > > public class MyHandler extends IoHandlerAdapter {
> > > IoSession mySession;
> > >
> > > public static void main(String[] args) throws Exception {
> > > IoAcceptor acceptor = new SocketAcceptor(2,
> > > Executors.newCachedThreadPool());
> > >
> > > ByteBuffer.setUseDirectBuffers(false);
> > > ByteBuffer.setAllocator(new SimpleByteBufferAllocator());
> > >
> > > // Prepare the configuration
> > > SocketAcceptorConfig cfg = new SocketAcceptorConfig();
> > > cfg.setReuseAddress(true);
> > > cfg.setThreadModel(ThreadModel.MANUAL);
> > > //cfg.getFilterChain().addLast("logger", new
> > LoggingFilter());
> > > cfg.getFilterChain().addLast(
> > > "codec",
> > > new ProtocolCodecFilter(new
> > TextLineCodecFactory(Charset
> > > .forName("UTF-8"))));
> > > cfg.getSessionConfig().setTcpNoDelay(true);
> > >
> > > // Bind
> > > acceptor.bind(new InetSocketAddress(8080),
> > > new MyHandler(), cfg);
> > >
> > > System.out.println("Listening on port 8080");
> > > }
> > >
> > >
> > > public void exceptionCaught(IoSession session, Throwable cause) {
> > > cause.printStackTrace();
> > > // Close connection when unexpected exception is caught.
> > > session.close();
> > > }
> > >
> > > public void sessionCreated(IoSession session) throws Exception {
> > > mySession = session;
> > >
> > > new MyThread().start();
> > > }
> > >
> > > private class MyThread extends Thread {
> > > public void run() {
> > > String message =
> > > "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
> > >
> > > while (true) {
> > >
> > > mySession.write(message);
> > > mySession.write(message);
> > > mySession.write(message);
> > >
> > > try {
> > > Thread.sleep(1);
> > > }
> > > catch (InterruptedException e) {
> > > // TODO Auto-generated catch block
> > > e.printStackTrace();
> > > }
> > > }
> > > }
> > > }
> > > }
> > >
> > > 2008/4/2 Steve Johns <steven.mark.johns@gmail.com
> > > <ma...@gmail.com>>:
> > >
> > > Actually I used windows console telnet command to connect to the
> > > server.
> > >
> > >
> > > On Wed, Apr 2, 2008 at 8:54 PM, "이희승 (Trustin Lee)
> > > <trustin@gmail.com <ma...@gmail.com>> wrote:
> > >
> > > Could you please upgrade to 1.1.6, 1.1.7-SNAPSHOT, 2.0.0-M1 or
> > > 2.0.0-M2-SNAPSHOT first?
> > >
> > > I don't think telnet client can be slow if it's implemented
> > > properly.
> > >
> > > Thanks,
> > >
> > > Steve Johns wrote:
> > > > Thank you very much.
> > > >
> > > > 1) I used Mina1.1.5 NOT 2.0 M1.
> > > > 2) Actually telnet client is on the same machine as my Mina
> > > server. Could it
> > > > be SLOW?
> > > >
> > > > In order to test how fast Mina could write, I modified the
> > > > ReverseProtocolHandler example. I added following code in
> > the
> > > > ReverseProtocolHandler:
> > > >
> > > > public void sessionCreated(IoSession session) throws
> > Exception {
> > > > String message =
> > > >
> > >
> > "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
> > > >
> > > > while (true) {
> > > > for (int i = 0; i < 5000; i++)
> > > > session.write(message);
> > > >
> > > > Thread.sleep(500);
> > > > }
> > > > }
> > > >
> > > > and I opened a telnet client connect to it. However, when I
> > > PrintGCDetails,
> > > > i obversed that GC is so busy to collect (take 0.15 sec to
> > > collect once, and
> > > > keep collecting )and after a while it seems too slow to
> > > collect. Eventually
> > > > OOM.
> > > >
> > > > I don't know if my code added was correct or NOT. I was
> > > impressed when you
> > > > said your server could handle 140k message/sec. How could
> > you
> > > possible do
> > > > that? ^^
> > > > On Wed, Apr 2, 2008 at 11:30 AM, "이희승 (Trustin Lee)
> > > <trustin@gmail.com <ma...@gmail.com>>
> > > > wrote:
> > > >
> > > >> Hi Steve,
> > > >>
> > > >> OOM can be raised when the write request rate is so high.
> > > There are two
> > > >> possible issues:
> > > >>
> > > >> 1) MINA NioProcessor is writing slowly.
> > > >> 2) Receiving party (i.e. telnet client) is receiving
> > slowly.
> > > >>
> > > >> If #1 is the case, we need to fix it. However, I was able
> > to
> > > achieve
> > > >> 140k messages per second for asynchronous messaging and 20k
> > > messages per
> > > >> second for synchronous messaging in my private performance
> > test
> > > >> performed in gigabit network recently. Just in case your
> > > scenario is
> > > >> triggering a hidden bug in MINA, please provide us a simple
> > test
> > > >> application.
> > > >>
> > > >> HTH,
> > > >>
> > > >> Steve Johns wrote:
> > > >>> So far Mina does a good job in my recent project until
> > > yesterday. I am
> > > >>> working a GW server handling 10k - 15k incoming
> > messages/sec
> > > with NOT
> > > >> very
> > > >>> big sizes and send them to clients. The weird thing is the
> > > program has
> > > >> NO
> > > >>> problem to consume the incoming messages(decoding, read).
> > > However, when
> > > >> a
> > > >>> simple telnet client connected to my server(same machine),
> > I
> > > observed
> > > >> the GC
> > > >>> taking 0.5 - 1.0sec to do GC every time and eventually
> > > OOM(GC is NOT
> > > >>> faster??). I got serveral questions:
> > > >>> 1) just wonder if mina is able to write that many of
> > > messages? I believe
> > > >>> telnet client on the same machine NOT considered to be a
> > > SLOW client,
> > > >> right?
> > > >>> 2) IF 1) Yes, the problem seems GC limitation but NOT Mina
> > > problem?
> > > >>>
> > > >>> Thanks.
> > > >>>
> > > >> --
> > > >> Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> > > >> --
> > > >> what we call human nature is actually human habit
> > > >> --
> > > >> http://gleamynode.net/
> > > >>
> > > >>
> > >
> > > --
> > > Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> > > --
> > > what we call human nature is actually human habit
> > > --
> > > http://gleamynode.net/
> > >
> > >
> > >
> >
> > --
> > Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> > --
> > what we call human nature is actually human habit
> > --
> > http://gleamynode.net/
> >
> >
>
Re: Write slow or OOM?
Posted by Steve Johns <st...@gmail.com>.
Thanks a lot.
Actually I fount out Thread.sleep(1) didn't work as what I expected. On
Windows XP, the minimum time Thread Acutally could sleep is 15ms. Which
meant my example didn't really simulate messages as much as expected. What I
tried to do was to generate 5k - 10k messages/sec and send them to the
client.
On Thu, Apr 3, 2008 at 5:55 PM, "이희승 (Trustin Lee) <tr...@gmail.com>
wrote:
> I was not able to reproduce OOM in my laptop (Intel Core 2 Duo T9300
> 2.5GHz 6MB L2 Cache, 2GB RAM, JDK 1.6.0.10-beta). The following is the
> source code I have used - it was simply ported to MINA 2:
>
> import java.net.InetSocketAddress;
> import java.nio.charset.Charset;
>
> import org.apache.mina.common.IoHandlerAdapter;
> import org.apache.mina.common.IoSession;
> import org.apache.mina.filter.codec.ProtocolCodecFilter;
> import org.apache.mina.filter.codec.textline.TextLineCodecFactory;
> import org.apache.mina.transport.socket.SocketAcceptor;
> import org.apache.mina.transport.socket.nio.NioSocketAcceptor;
>
> public class MyHandler extends IoHandlerAdapter {
> IoSession mySession;
>
> public static void main(String[] args) throws Exception {
> SocketAcceptor acceptor = new NioSocketAcceptor(2);
>
> // Prepare the configuration
> acceptor.getFilterChain().addLast(
> "codec",
> new ProtocolCodecFilter(new TextLineCodecFactory(Charset
> .forName("UTF-8"))));
> acceptor.getSessionConfig().setTcpNoDelay(true);
>
> // Bind
> acceptor.setHandler(new MyHandler());
> acceptor.bind(new InetSocketAddress(8080));
>
> System.out.println("Listening on port 8080");
> }
>
> @Override
> public void exceptionCaught(IoSession session, Throwable cause) {
> cause.printStackTrace();
> // Close connection when unexpected exception is caught.
> session.close();
> }
>
> @Override
> public void sessionCreated(IoSession session) throws Exception {
> mySession = session;
>
> new MyThread().start();
> }
>
> private class MyThread extends Thread {
> @Override
> public void run() {
> String message =
> "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
>
> while (true) {
>
> mySession.write(message);
> mySession.write(message);
> mySession.write(message);
>
> try {
> Thread.sleep(1);
> } catch (InterruptedException e) {
> // TODO Auto-generated catch block
> e.printStackTrace();
> }
> }
> }
> }
> }
>
> On the client side, I used netcat:
>
> $ nc localhost 8080
>
> Please note my test was run in Linux. I'm not sure about Windows
> unfortunately. There's also netcat for Windows IIRC - you could try
> that instead of Windows telnet, which doesn't seem to perform very well.
>
> HTH,
>
> Steve Johns wrote:
> > I wrote the following testing program. I used one thread to simulate the
> > incoming datafeed. Sleep(1) = 1000messages/sec(I added 3
> > session.write(message)), then 3000 messages / sec roughly sent to
> client.
> > My server CPU reaches 70% with one telnet client on the same machine.
> >
> -----------------------------------------------------------------------------------
> > public class MyHandler extends IoHandlerAdapter {
> > IoSession mySession;
> >
> > public static void main(String[] args) throws Exception {
> > IoAcceptor acceptor = new SocketAcceptor(2,
> > Executors.newCachedThreadPool());
> >
> > ByteBuffer.setUseDirectBuffers(false);
> > ByteBuffer.setAllocator(new SimpleByteBufferAllocator());
> >
> > // Prepare the configuration
> > SocketAcceptorConfig cfg = new SocketAcceptorConfig();
> > cfg.setReuseAddress(true);
> > cfg.setThreadModel(ThreadModel.MANUAL);
> > //cfg.getFilterChain().addLast("logger", new LoggingFilter());
> > cfg.getFilterChain().addLast(
> > "codec",
> > new ProtocolCodecFilter(new
> TextLineCodecFactory(Charset
> > .forName("UTF-8"))));
> > cfg.getSessionConfig().setTcpNoDelay(true);
> >
> > // Bind
> > acceptor.bind(new InetSocketAddress(8080),
> > new MyHandler(), cfg);
> >
> > System.out.println("Listening on port 8080");
> > }
> >
> >
> > public void exceptionCaught(IoSession session, Throwable cause) {
> > cause.printStackTrace();
> > // Close connection when unexpected exception is caught.
> > session.close();
> > }
> >
> > public void sessionCreated(IoSession session) throws Exception {
> > mySession = session;
> >
> > new MyThread().start();
> > }
> >
> > private class MyThread extends Thread {
> > public void run() {
> > String message =
> > "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
> >
> > while (true) {
> >
> > mySession.write(message);
> > mySession.write(message);
> > mySession.write(message);
> >
> > try {
> > Thread.sleep(1);
> > }
> > catch (InterruptedException e) {
> > // TODO Auto-generated catch block
> > e.printStackTrace();
> > }
> > }
> > }
> > }
> > }
> >
> > 2008/4/2 Steve Johns <steven.mark.johns@gmail.com
> > <ma...@gmail.com>>:
> >
> > Actually I used windows console telnet command to connect to the
> > server.
> >
> >
> > On Wed, Apr 2, 2008 at 8:54 PM, "이희승 (Trustin Lee)
> > <trustin@gmail.com <ma...@gmail.com>> wrote:
> >
> > Could you please upgrade to 1.1.6, 1.1.7-SNAPSHOT, 2.0.0-M1 or
> > 2.0.0-M2-SNAPSHOT first?
> >
> > I don't think telnet client can be slow if it's implemented
> > properly.
> >
> > Thanks,
> >
> > Steve Johns wrote:
> > > Thank you very much.
> > >
> > > 1) I used Mina1.1.5 NOT 2.0 M1.
> > > 2) Actually telnet client is on the same machine as my Mina
> > server. Could it
> > > be SLOW?
> > >
> > > In order to test how fast Mina could write, I modified the
> > > ReverseProtocolHandler example. I added following code in the
> > > ReverseProtocolHandler:
> > >
> > > public void sessionCreated(IoSession session) throws Exception
> {
> > > String message =
> > >
> >
> "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
> > >
> > > while (true) {
> > > for (int i = 0; i < 5000; i++)
> > > session.write(message);
> > >
> > > Thread.sleep(500);
> > > }
> > > }
> > >
> > > and I opened a telnet client connect to it. However, when I
> > PrintGCDetails,
> > > i obversed that GC is so busy to collect (take 0.15 sec to
> > collect once, and
> > > keep collecting )and after a while it seems too slow to
> > collect. Eventually
> > > OOM.
> > >
> > > I don't know if my code added was correct or NOT. I was
> > impressed when you
> > > said your server could handle 140k message/sec. How could you
> > possible do
> > > that? ^^
> > > On Wed, Apr 2, 2008 at 11:30 AM, "이희승 (Trustin Lee)
> > <trustin@gmail.com <ma...@gmail.com>>
> > > wrote:
> > >
> > >> Hi Steve,
> > >>
> > >> OOM can be raised when the write request rate is so high.
> > There are two
> > >> possible issues:
> > >>
> > >> 1) MINA NioProcessor is writing slowly.
> > >> 2) Receiving party (i.e. telnet client) is receiving slowly.
> > >>
> > >> If #1 is the case, we need to fix it. However, I was able to
> > achieve
> > >> 140k messages per second for asynchronous messaging and 20k
> > messages per
> > >> second for synchronous messaging in my private performance
> test
> > >> performed in gigabit network recently. Just in case your
> > scenario is
> > >> triggering a hidden bug in MINA, please provide us a simple
> test
> > >> application.
> > >>
> > >> HTH,
> > >>
> > >> Steve Johns wrote:
> > >>> So far Mina does a good job in my recent project until
> > yesterday. I am
> > >>> working a GW server handling 10k - 15k incoming messages/sec
> > with NOT
> > >> very
> > >>> big sizes and send them to clients. The weird thing is the
> > program has
> > >> NO
> > >>> problem to consume the incoming messages(decoding, read).
> > However, when
> > >> a
> > >>> simple telnet client connected to my server(same machine), I
> > observed
> > >> the GC
> > >>> taking 0.5 - 1.0sec to do GC every time and eventually
> > OOM(GC is NOT
> > >>> faster??). I got serveral questions:
> > >>> 1) just wonder if mina is able to write that many of
> > messages? I believe
> > >>> telnet client on the same machine NOT considered to be a
> > SLOW client,
> > >> right?
> > >>> 2) IF 1) Yes, the problem seems GC limitation but NOT Mina
> > problem?
> > >>>
> > >>> Thanks.
> > >>>
> > >> --
> > >> Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> > >> --
> > >> what we call human nature is actually human habit
> > >> --
> > >> http://gleamynode.net/
> > >>
> > >>
> >
> > --
> > Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> > --
> > what we call human nature is actually human habit
> > --
> > http://gleamynode.net/
> >
> >
> >
>
> --
> Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> --
> what we call human nature is actually human habit
> --
> http://gleamynode.net/
>
>
Re: Write slow or OOM?
Posted by 이희승 (Trustin Lee) <t...@gmail.com>.
I was not able to reproduce OOM in my laptop (Intel Core 2 Duo T9300
2.5GHz 6MB L2 Cache, 2GB RAM, JDK 1.6.0.10-beta). The following is the
source code I have used - it was simply ported to MINA 2:
import java.net.InetSocketAddress;
import java.nio.charset.Charset;
import org.apache.mina.common.IoHandlerAdapter;
import org.apache.mina.common.IoSession;
import org.apache.mina.filter.codec.ProtocolCodecFilter;
import org.apache.mina.filter.codec.textline.TextLineCodecFactory;
import org.apache.mina.transport.socket.SocketAcceptor;
import org.apache.mina.transport.socket.nio.NioSocketAcceptor;
public class MyHandler extends IoHandlerAdapter {
IoSession mySession;
public static void main(String[] args) throws Exception {
SocketAcceptor acceptor = new NioSocketAcceptor(2);
// Prepare the configuration
acceptor.getFilterChain().addLast(
"codec",
new ProtocolCodecFilter(new TextLineCodecFactory(Charset
.forName("UTF-8"))));
acceptor.getSessionConfig().setTcpNoDelay(true);
// Bind
acceptor.setHandler(new MyHandler());
acceptor.bind(new InetSocketAddress(8080));
System.out.println("Listening on port 8080");
}
@Override
public void exceptionCaught(IoSession session, Throwable cause) {
cause.printStackTrace();
// Close connection when unexpected exception is caught.
session.close();
}
@Override
public void sessionCreated(IoSession session) throws Exception {
mySession = session;
new MyThread().start();
}
private class MyThread extends Thread {
@Override
public void run() {
String message =
"aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
while (true) {
mySession.write(message);
mySession.write(message);
mySession.write(message);
try {
Thread.sleep(1);
} catch (InterruptedException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
}
}
}
On the client side, I used netcat:
$ nc localhost 8080
Please note my test was run in Linux. I'm not sure about Windows
unfortunately. There's also netcat for Windows IIRC - you could try
that instead of Windows telnet, which doesn't seem to perform very well.
HTH,
Steve Johns wrote:
> I wrote the following testing program. I used one thread to simulate the
> incoming datafeed. Sleep(1) = 1000messages/sec(I added 3
> session.write(message)), then 3000 messages / sec roughly sent to client.
> My server CPU reaches 70% with one telnet client on the same machine.
> -----------------------------------------------------------------------------------
> public class MyHandler extends IoHandlerAdapter {
> IoSession mySession;
>
> public static void main(String[] args) throws Exception {
> IoAcceptor acceptor = new SocketAcceptor(2,
> Executors.newCachedThreadPool());
>
> ByteBuffer.setUseDirectBuffers(false);
> ByteBuffer.setAllocator(new SimpleByteBufferAllocator());
>
> // Prepare the configuration
> SocketAcceptorConfig cfg = new SocketAcceptorConfig();
> cfg.setReuseAddress(true);
> cfg.setThreadModel(ThreadModel.MANUAL);
> //cfg.getFilterChain().addLast("logger", new LoggingFilter());
> cfg.getFilterChain().addLast(
> "codec",
> new ProtocolCodecFilter(new TextLineCodecFactory(Charset
> .forName("UTF-8"))));
> cfg.getSessionConfig().setTcpNoDelay(true);
>
> // Bind
> acceptor.bind(new InetSocketAddress(8080),
> new MyHandler(), cfg);
>
> System.out.println("Listening on port 8080");
> }
>
>
> public void exceptionCaught(IoSession session, Throwable cause) {
> cause.printStackTrace();
> // Close connection when unexpected exception is caught.
> session.close();
> }
>
> public void sessionCreated(IoSession session) throws Exception {
> mySession = session;
>
> new MyThread().start();
> }
>
> private class MyThread extends Thread {
> public void run() {
> String message =
> "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
>
> while (true) {
>
> mySession.write(message);
> mySession.write(message);
> mySession.write(message);
>
> try {
> Thread.sleep(1);
> }
> catch (InterruptedException e) {
> // TODO Auto-generated catch block
> e.printStackTrace();
> }
> }
> }
> }
> }
>
> 2008/4/2 Steve Johns <steven.mark.johns@gmail.com
> <ma...@gmail.com>>:
>
> Actually I used windows console telnet command to connect to the
> server.
>
>
> On Wed, Apr 2, 2008 at 8:54 PM, "이희승 (Trustin Lee)
> <trustin@gmail.com <ma...@gmail.com>> wrote:
>
> Could you please upgrade to 1.1.6, 1.1.7-SNAPSHOT, 2.0.0-M1 or
> 2.0.0-M2-SNAPSHOT first?
>
> I don't think telnet client can be slow if it's implemented
> properly.
>
> Thanks,
>
> Steve Johns wrote:
> > Thank you very much.
> >
> > 1) I used Mina1.1.5 NOT 2.0 M1.
> > 2) Actually telnet client is on the same machine as my Mina
> server. Could it
> > be SLOW?
> >
> > In order to test how fast Mina could write, I modified the
> > ReverseProtocolHandler example. I added following code in the
> > ReverseProtocolHandler:
> >
> > public void sessionCreated(IoSession session) throws Exception {
> > String message =
> >
> "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
> >
> > while (true) {
> > for (int i = 0; i < 5000; i++)
> > session.write(message);
> >
> > Thread.sleep(500);
> > }
> > }
> >
> > and I opened a telnet client connect to it. However, when I
> PrintGCDetails,
> > i obversed that GC is so busy to collect (take 0.15 sec to
> collect once, and
> > keep collecting )and after a while it seems too slow to
> collect. Eventually
> > OOM.
> >
> > I don't know if my code added was correct or NOT. I was
> impressed when you
> > said your server could handle 140k message/sec. How could you
> possible do
> > that? ^^
> > On Wed, Apr 2, 2008 at 11:30 AM, "이희승 (Trustin Lee)
> <trustin@gmail.com <ma...@gmail.com>>
> > wrote:
> >
> >> Hi Steve,
> >>
> >> OOM can be raised when the write request rate is so high.
> There are two
> >> possible issues:
> >>
> >> 1) MINA NioProcessor is writing slowly.
> >> 2) Receiving party (i.e. telnet client) is receiving slowly.
> >>
> >> If #1 is the case, we need to fix it. However, I was able to
> achieve
> >> 140k messages per second for asynchronous messaging and 20k
> messages per
> >> second for synchronous messaging in my private performance test
> >> performed in gigabit network recently. Just in case your
> scenario is
> >> triggering a hidden bug in MINA, please provide us a simple test
> >> application.
> >>
> >> HTH,
> >>
> >> Steve Johns wrote:
> >>> So far Mina does a good job in my recent project until
> yesterday. I am
> >>> working a GW server handling 10k - 15k incoming messages/sec
> with NOT
> >> very
> >>> big sizes and send them to clients. The weird thing is the
> program has
> >> NO
> >>> problem to consume the incoming messages(decoding, read).
> However, when
> >> a
> >>> simple telnet client connected to my server(same machine), I
> observed
> >> the GC
> >>> taking 0.5 - 1.0sec to do GC every time and eventually
> OOM(GC is NOT
> >>> faster??). I got serveral questions:
> >>> 1) just wonder if mina is able to write that many of
> messages? I believe
> >>> telnet client on the same machine NOT considered to be a
> SLOW client,
> >> right?
> >>> 2) IF 1) Yes, the problem seems GC limitation but NOT Mina
> problem?
> >>>
> >>> Thanks.
> >>>
> >> --
> >> Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> >> --
> >> what we call human nature is actually human habit
> >> --
> >> http://gleamynode.net/
> >>
> >>
>
> --
> Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> --
> what we call human nature is actually human habit
> --
> http://gleamynode.net/
>
>
>
--
Trustin Lee - Principal Software Engineer, JBoss, Red Hat
--
what we call human nature is actually human habit
--
http://gleamynode.net/
Re: Write slow or OOM?
Posted by 이희승 (Trustin Lee) <t...@gmail.com>.
Could you please upgrade to 1.1.6, 1.1.7-SNAPSHOT, 2.0.0-M1 or
2.0.0-M2-SNAPSHOT first?
I don't think telnet client can be slow if it's implemented properly.
Thanks,
Steve Johns wrote:
> Thank you very much.
>
> 1) I used Mina1.1.5 NOT 2.0 M1.
> 2) Actually telnet client is on the same machine as my Mina server. Could it
> be SLOW?
>
> In order to test how fast Mina could write, I modified the
> ReverseProtocolHandler example. I added following code in the
> ReverseProtocolHandler:
>
> public void sessionCreated(IoSession session) throws Exception {
> String message =
> "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
>
> while (true) {
> for (int i = 0; i < 5000; i++)
> session.write(message);
>
> Thread.sleep(500);
> }
> }
>
> and I opened a telnet client connect to it. However, when I PrintGCDetails,
> i obversed that GC is so busy to collect (take 0.15 sec to collect once, and
> keep collecting )and after a while it seems too slow to collect. Eventually
> OOM.
>
> I don't know if my code added was correct or NOT. I was impressed when you
> said your server could handle 140k message/sec. How could you possible do
> that? ^^
> On Wed, Apr 2, 2008 at 11:30 AM, "이희승 (Trustin Lee) <tr...@gmail.com>
> wrote:
>
>> Hi Steve,
>>
>> OOM can be raised when the write request rate is so high. There are two
>> possible issues:
>>
>> 1) MINA NioProcessor is writing slowly.
>> 2) Receiving party (i.e. telnet client) is receiving slowly.
>>
>> If #1 is the case, we need to fix it. However, I was able to achieve
>> 140k messages per second for asynchronous messaging and 20k messages per
>> second for synchronous messaging in my private performance test
>> performed in gigabit network recently. Just in case your scenario is
>> triggering a hidden bug in MINA, please provide us a simple test
>> application.
>>
>> HTH,
>>
>> Steve Johns wrote:
>>> So far Mina does a good job in my recent project until yesterday. I am
>>> working a GW server handling 10k - 15k incoming messages/sec with NOT
>> very
>>> big sizes and send them to clients. The weird thing is the program has
>> NO
>>> problem to consume the incoming messages(decoding, read). However, when
>> a
>>> simple telnet client connected to my server(same machine), I observed
>> the GC
>>> taking 0.5 - 1.0sec to do GC every time and eventually OOM(GC is NOT
>>> faster??). I got serveral questions:
>>> 1) just wonder if mina is able to write that many of messages? I believe
>>> telnet client on the same machine NOT considered to be a SLOW client,
>> right?
>>> 2) IF 1) Yes, the problem seems GC limitation but NOT Mina problem?
>>>
>>> Thanks.
>>>
>> --
>> Trustin Lee - Principal Software Engineer, JBoss, Red Hat
>> --
>> what we call human nature is actually human habit
>> --
>> http://gleamynode.net/
>>
>>
--
Trustin Lee - Principal Software Engineer, JBoss, Red Hat
--
what we call human nature is actually human habit
--
http://gleamynode.net/
Re: Write slow or OOM?
Posted by Steve Johns <st...@gmail.com>.
Thank you very much.
1) I used Mina1.1.5 NOT 2.0 M1.
2) Actually telnet client is on the same machine as my Mina server. Could it
be SLOW?
In order to test how fast Mina could write, I modified the
ReverseProtocolHandler example. I added following code in the
ReverseProtocolHandler:
public void sessionCreated(IoSession session) throws Exception {
String message =
"aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa";
while (true) {
for (int i = 0; i < 5000; i++)
session.write(message);
Thread.sleep(500);
}
}
and I opened a telnet client connect to it. However, when I PrintGCDetails,
i obversed that GC is so busy to collect (take 0.15 sec to collect once, and
keep collecting )and after a while it seems too slow to collect. Eventually
OOM.
I don't know if my code added was correct or NOT. I was impressed when you
said your server could handle 140k message/sec. How could you possible do
that? ^^
On Wed, Apr 2, 2008 at 11:30 AM, "이희승 (Trustin Lee) <tr...@gmail.com>
wrote:
> Hi Steve,
>
> OOM can be raised when the write request rate is so high. There are two
> possible issues:
>
> 1) MINA NioProcessor is writing slowly.
> 2) Receiving party (i.e. telnet client) is receiving slowly.
>
> If #1 is the case, we need to fix it. However, I was able to achieve
> 140k messages per second for asynchronous messaging and 20k messages per
> second for synchronous messaging in my private performance test
> performed in gigabit network recently. Just in case your scenario is
> triggering a hidden bug in MINA, please provide us a simple test
> application.
>
> HTH,
>
> Steve Johns wrote:
> > So far Mina does a good job in my recent project until yesterday. I am
> > working a GW server handling 10k - 15k incoming messages/sec with NOT
> very
> > big sizes and send them to clients. The weird thing is the program has
> NO
> > problem to consume the incoming messages(decoding, read). However, when
> a
> > simple telnet client connected to my server(same machine), I observed
> the GC
> > taking 0.5 - 1.0sec to do GC every time and eventually OOM(GC is NOT
> > faster??). I got serveral questions:
> > 1) just wonder if mina is able to write that many of messages? I believe
> > telnet client on the same machine NOT considered to be a SLOW client,
> right?
> > 2) IF 1) Yes, the problem seems GC limitation but NOT Mina problem?
> >
> > Thanks.
> >
>
> --
> Trustin Lee - Principal Software Engineer, JBoss, Red Hat
> --
> what we call human nature is actually human habit
> --
> http://gleamynode.net/
>
>
Re: Write slow or OOM?
Posted by 이희승 (Trustin Lee) <t...@gmail.com>.
Hi Steve,
OOM can be raised when the write request rate is so high. There are two
possible issues:
1) MINA NioProcessor is writing slowly.
2) Receiving party (i.e. telnet client) is receiving slowly.
If #1 is the case, we need to fix it. However, I was able to achieve
140k messages per second for asynchronous messaging and 20k messages per
second for synchronous messaging in my private performance test
performed in gigabit network recently. Just in case your scenario is
triggering a hidden bug in MINA, please provide us a simple test
application.
HTH,
Steve Johns wrote:
> So far Mina does a good job in my recent project until yesterday. I am
> working a GW server handling 10k - 15k incoming messages/sec with NOT very
> big sizes and send them to clients. The weird thing is the program has NO
> problem to consume the incoming messages(decoding, read). However, when a
> simple telnet client connected to my server(same machine), I observed the GC
> taking 0.5 - 1.0sec to do GC every time and eventually OOM(GC is NOT
> faster??). I got serveral questions:
> 1) just wonder if mina is able to write that many of messages? I believe
> telnet client on the same machine NOT considered to be a SLOW client, right?
> 2) IF 1) Yes, the problem seems GC limitation but NOT Mina problem?
>
> Thanks.
>
--
Trustin Lee - Principal Software Engineer, JBoss, Red Hat
--
what we call human nature is actually human habit
--
http://gleamynode.net/