TC65 : Slow UDP reception

Receiving UDP datagrams on the TC65 is easy but inefficient. You need to create a new thread if you want to receive data asynchronously. There’s no way of knowing if a new datagram has been received or not without hanging on the ::receive( Datagram ); method. This is quite weird considering you can do it in TCP.

The real problem is it’s freaking slow. The Cinterion documentation just tells you “This method blocks until a datagram is received.”. What it doesn’t say is that nearly each time, it blocks for 100 to 700 ms after UDP datagrams have actually been received. And worse, this slowness/sleeping avoids the program from treating data and finally throws a little java.io.IOException exception : “No buffer space available”. I tried to give the udp receiving thread a higher priority than the other threads, I tried to make the TCP reception thread sleep a lot just in case it would lock some kind of network object, I looked on how could this method be implemented but couldn’t find a solution and don’t believe there’s one.

By the way, most of the codes on UDP data reception are lame. They all show the same stupid synchronous code. The real code you will need is something like that :

You can remove the Logger class calls. It enables me to make some build-specific (mostly logging) code. The “if” false conditions are remove by the java compiler (which is actually mostly a pre-compiler). I will (quickly this time) talk about this someday…

This program isn’t TC65 specific, it should work with any MIDP enabled J2ME device.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
package Network.UDPReceiver;
 
import Common.Logger;
import java.io.IOException;
import javax.microedition.io.Datagram;
import javax.microedition.io.DatagramConnection;
 
/**
 * UDP event-like receiving class
 * @author Florent Clairambault
 */
public class UDPReceiver implements Runnable {
 
    private IUDPReceive _receivingClass;
    private DatagramConnection _udpConn;
    private Datagram _udpDatagram;
    private Thread _thread;
    private boolean _loop;
 
    /**
     * Creates and launchs the receiver
     * @param conn The connection
     * @param receivingClass The receiving class of the UDP receive event
     */
    public UDPReceiver(DatagramConnection conn, IUDPReceive receivingClass) {
        _udpConn = conn;
        _receivingClass = receivingClass;
        init();
        start();
    }
 
    /**
     * Prepares the datagram used to receive data
     */
    private void init() {
        if (Logger.E_VERBOSE)
            Logger.Log( 32, "UDPReceiver.init();" );
        try {
            int maxLength = _udpConn.getMaximumLength();
            if (Logger.E_DEBUG) {
                Logger.Log( 39, "maxLength=" + maxLength );
            }
            _udpDatagram = _udpConn.newDatagram( maxLength );
        } catch (IOException ex) {
 
            // This should NEVER happen !
            if (Logger.E_CRITICAL)
                Logger.Log( 37, "UDPReceiver.init", ex );
        }
    }
 
    /**
     * The actual data reception
     * @return the data received
     * @throws java.io.IOException When something fais, it means we have to stop
     */
    public synchronized byte[] receiveUdpFrame() throws IOException {
        if (Logger.E_DEBUG)
            Logger.Log( 46, "UDPReceiver.receiveUdpFrame();" );
 
        _udpConn.receive( _udpDatagram );
        int size = _udpDatagram.getLength();
        if (Logger.E_DEBUG)
            Logger.Log( 49, "size = " + size );
 
        // We copy the data so that it can be used on an other thread
        byte[] data = new byte[size];
        System.arraycopy( _udpDatagram.getData(), 0, data, 0, size );
 
        // These two lines might seem weird but are the more efficient way
        // to prepare next datagram reception.
        _udpDatagram.reset();
        _udpDatagram.setLength( _udpDatagram.getData().length );
 
        if (Logger.E_DEBUG)
            Logger.Log( 63, "UDPReceiver.receiveUdpFrame : ok !" );
 
        return data;
    }
 
    /**
     * Starts the receiving thread
     */
    private void start() {
        _loop = true;
 
        _thread = new Thread( this, "udp" );
        _thread.start();
    }
 
    /**
     * Plan to stop
     * 
     * This method is in fact useless. The thread will stop when the
     * DatagramConnection (_udpConn) will be closed.
     */
    public void stop() {
        _loop = false;
    }
 
    /**
     * The thread method
     */
    public void run() {
        try {
            while (_loop) {
                if (Logger.E_DEBUG)
                    Logger.Log( 70, "UDPReceiver : running..." );
 
                byte[] data = receiveUdpFrame();
 
                // We "throw" an event-like method
                _receivingClass.UdpDataReceived( data );
            }
        } catch (IOException ex) {
            if (Logger.E_DEBUG)
                Logger.Log( 54, "UDPReceiver.run", ex );
 
            // The connection must have been closed, we have to stop !
            _loop = false;
        } catch (Exception ex) {
            if (Logger.E_CRITICAL)
                Logger.Log( 119, "UDPReceiver.run", ex );
        }
    }
}

The “mother” class has implements this IUDPReceiver interface :

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
package Network.UDPReceiver;
 
/**
 * Interface an asynchronous UDP receiving class must implement
 * @author Florent Clairambault
 */
public interface IUDPReceive {
 
    /**
     * Called when a datagram is received on a connection
     * @param data Content of the datagram
     * 
     * We assume that the receiving class already knows the sender of the
     * datagram because it's a connection established with an "UDP server".
     */
    void UdpDataReceived( byte[] data );
}

The “mother” class should open the connection like that :

1
2
_udpConn = (UDPDatagramConnection) Connector.open( "datagram://" + server );
_udpReceiver = new UDPReceiver( _udpConn, this );

where server is an adresse like that “x.x.x.x:y”. It can also be used as a UDP server with “:y”.

And has to implement the interface IUDPReceiver interface.

Note on receiving UDP over GPRS

Most GPRS connections (every one of them unless you subscribe to special options), are “NATed” (using a NAT router). They are behind a private network (10.x.x.x/8, 192.168.x.x/16, 172.16.x.x/12). That prevents the equipments from being directly accessible (by other means than their phone number).

On TCP, you have to establish a connection with a remote host (that you’re quite likely to call a “server”), and then the server can answer on the link. You have a reliable bi-directional stream you can use to transmit anything you’d like. You can keep the connection open running forever, you just have to send data or set the KeepAlive option. NAT routers break TCP connections when nothing has been transmitted on it for 1 hour to 4 days. This has even been considered like a “bug” by some people because the NAT router reach its memory limit or the max number of TCP connections it can translate (from private to public).
On the TC65, it means you can communicate over a TCP connections for months. You just have to make sure that it’s still alive by providing a little ping mechanism in your protocol (you can send pings every 15/30 minutes for instance).

On UDP, it’s roughly the same thing except there’s no connection. In fact, UDP is minimalist (it only has a 8 bytes header). You have no mechanism for connection establishment, you just throw your data (and by chance, it will be received). The router can’t possibly know when the transmission has ended. So when the router supports bi-directional UDP, it just tries to keep track of the last connections established for a limited time. My personal tomato firmware enabled router, keeps track of the connections when you sent data from LAN to WAN in the last 30 seconds, and that goes to 180 seconds when the other side replies.

So, basically what you need to do to set up a bi-directionnal UDP connection is sending some data from the equipment to the server each 25 seconds. My guess is that even empty UDP packets (8 bytes sized packet) should work.

The most simple/secure way to do this (don’t forget that UDP datagrams can be dropped on their way to the server), is to establish a communication with the server in TCP. Then, ask for a unique identifier (two bytes for instance) and then use it each time you send data on UDP.

If you want to send data between two NATed network, it’s also possible. This little RFC explains it very well. If this seems a little bit to abstract, you should try out wireshark. I always use it when I have a doubts about some low-level network transmission (I like to investigate a lot, just to make sure I’m not missing anything).

I didn’t say it sucks

It didn’t say the TC65 sucks. It’s just not as efficient as I thought it could be. The network stack doesn’t seem to have been built for performances. If you send TCP data or receive UDP data too often, there’s a pretty good chance you will face the same problems. If you found a solution to bypass these problems, please tell me how.

But remember, if you only need to send some data each second (or less often), which is the case in most of the M2M applications, you can be sure the chip will suit your needs.

3 thoughts on “TC65 : Slow UDP reception”

  1. Hi,

    You can disable the reverse DNS lookup in TC65 Rel 3 and higher.
    try
    {
    UDPDatagramConnection UDatagram = null;
    UDatagram= (UDPDatagramConnection) Connector.open(“datagram://host:7;bearer_type=gprs;access_point=APN_NAME;timeout=30;reverse_dns_lookup=no”);

    }
    catch (IOException e)
    {
    e.printStackTrace();
    }

    Other way is to put and real DNS in the configuration.

    Regards

    Pua

  2. So, your saying that any received UDP packet will launch a reverse DNS request. It seems a little weird since it sometimes only blocked for 100/200 ms.

    Anyway, the chip I was working with was a TC65 v2. But that’s a good thing to know. I might do some other performance tests with this setting configured.

    Thank you very much for this hint.

  3. Slow UDP receive….

    I had this once. I cannot recall which of these 2 changes solved it, but it did provide me with “immediate” UDP receive, instead of “some seconds”

    1. If the connection is declared as UDPDatagramConnection for the sender, also declare the same for the receiver (ie. not just DatagramConnection)

    2. Instead of receiving the datagram into the “internal” bytearray, delcare your own. Something like…

    byte[] buf = new byte[256];
    Datagram datagram = connection.newDatagram(buf, buf.length);
    datagram.setLength(buf.length);
    connection.receive(datagram);
    System.out.println(“Datagram received: ” + buf.toString() + “, ” + datagram.getLength() + ” bytes”;

    May or may not be the same issue, but just in case it might help! Regards.

Leave a Reply

Your email address will not be published. Required fields are marked *