Creating a Custom Socket Type

Many times, it is necessary to process data before sending it to, or after receiving it from, a socket. Instead of using java.net.Socket sockets with pre- and post-processing, another type of socket can be used.

This page shows you the steps to follow to create a custom socket type. It also contains an example implementation of a custom socket named CompressionSocket, which compresses data sent over the connection it uses.

The CompressionSocket and its suppporting classes are also used in our tutorial on how to create a custom RMI socket factory. As a result, the source files for this example are all in the examples.rmisocfac package.

In this tutorial, creating the CompressionSocket class and related classes is broken down into the following four steps, which can be adapted to create other custom socket type:

1. Extend java.io.FilterOutputStream to create an output stream for the socket. Override FilterOutputStream methods as necessary.
2. Extend java.io.FilterInputStream to create an input stream for the socket. Override FilterInputStream methods as necessary.
3. Create a java.net.Socket subclass. Implement the appropriate constructors and override the getInputStream, getOutputStream and close methods.
4. Create a java.net.ServerSocket subclass. Implement the constructor, and overload the accept method to create a socket of the desired type.

Since we are writing a socket that does data compression, we will write a class named CompressionOutputStream that extends FilterOutputStream. However, extending FilterOutputStream will not be appropriate in every case. In general, you should extend the type of output stream best-suited for the type of socket you are implementing. In this example, FilterOutputStream is most appropriate.

For our example, we will use a very simple algorithm which provides 6-bit character encoding for up to 62 common characters and 18-bit encoding for the other characters (remember that in Java characters are usually 16 bits). There is a lookup table to map the 62 common characters to 6-bit numbers in the range 0 to 61. Depending on the result of the lookup, each encoded character is marked with a constant to indicate whether it has 6-bit or 18-bit encoding. Finally, the encoded characters are written to the stream. This algorithm assumes that all characters encountered are ASCII, and that the high-order byte of each character is unused.

Please note: this algorithm is not recommended for use in any application requiring data compression. It is included only for this example and is not intended for practical use.

Before writing any sample code demonstrating how to create custom sockets, it is necessary to write an interface containing the information to be shared between the input and output stream classes. For this algorithm, the common character lookup table, codeTable and 3 constants are included as members. It is useful to note that while an interface is included in this example, an interface is not always required.

interface CompressionConstants {
 
    /** Constants for 6-bit code values. */

    /** No operation: used to pad words on flush. */
    static final int NOP     = 0;  

    /** Introduces raw byte format. */
    static final int RAW     = 1;  

    /** Format indicator for characters found in lookup table. */
    static final int BASE    = 2;  

    /** A character's code is it's index in the lookup table. */
    static final String codeTable =
        "abcdefghijklmnopqrstuvwxyz" +
        "ABCDEFGHIJKLMNOPQRSTUVWXYZ ,.!?\"'()";
}
	

The source code for the above interface can be found here.

Now that the interface for sharing information between the input and output streams has been defined, we will complete step 1, which was to write a class that extends java.io.FilterOutputStream to create an output stream for the socket, overriding FilterOutputStream methods as necessary.

Below is the source code for the class CompressionOutputStream.java. An explanation of the compression algorithm can be found in the comments of the source code. An explanation of the implementation follows the code.

import java.io.*;
 
class CompressionOutputStream extends FilterOutputStream
    implements CompressionConstants
{

    /*
     * Constructor calls constructor of superclass.
     */
    public CompressionOutputStream(OutputStream out) {
        super(out);
    }
 
    /* 
     * Buffer of 6-bit codes to pack into next 32-bit
     * word.  Five 6-bit codes fit into 4 words. 
     */
    int buf[] = new int[5];
 
    /*
     * Index of valid codes waiting in buf. 
     */
    int bufPos = 0;
 

    /*  
     * This method writes one byte to the socket stream. 
     */ 
    public void write(int b) throws IOException {
        // force argument to one byte
        b &= 0xFF;                      
 
        // Look up pos in codeTable to get its encoding. 
        int pos = codeTable.indexOf((char)b);

        if (pos != -1)
            // If pos is in the codeTable, write 
            // BASE + pos into buf. By adding BASE 
            // to pos, we know that the characters in
            // the codeTable will always have a code 
            // between 2 and 63 inclusive. This allows 
            // us to use RAW (RAW is equal to 1) to signify 
            // that the next two groups of 6-bits are necessary
            // for decompression of the next character. 

            writeCode(BASE + pos);

        else {
            // Otherwise, write RAW into buf to signify that
            // the Character is being sent in 12 bits.
            writeCode(RAW);

            // Write the last 4 bits of b into the buf.
            writeCode(b >> 4);

            // Truncate b to contain data in only the first 4
            // bits and write the first 4 bits of b into buf.
            writeCode(b & 0xF);
        }
    }
 
    /* 
     * This method writes up to len bytes to the socket stream. 
     */
    public void write(byte b[], int off, int len) 
        throws IOException 
    {
        // This implementation is quite inefficient because 
        // it has to call the other write method for every 
        // byte in the array.  It could be optimized for 
        // performance by doing all the processing in this 
        // method.

        for (int i = 0; i < len; i++)
            write(b[off + i]);
    }
 
   /* 
    * Clears buffer of all data (zeroes it out). 
    */ 
   public void flush() throws IOException {
        while (bufPos > 0)
            writeCode(NOP);
    }
 
    /* 
     * This method actually puts the data into the output stream
     * after packing the data from all 5 bytes in buf into one
     * word. Remember, each byte has, at most, 6 significant bits.
     */
    private void writeCode(int c) throws IOException {
        buf[bufPos++] = c;

        // write next word when we have 5 codes
        if (bufPos == 5) {      
            int pack = (buf[0] << 24) | (buf[1] << 18) | 
                       (buf[2] << 12) | (buf[3] << 6) | buf[4];
            out.write((pack >>> 24) & 0xFF);
            out.write((pack >>> 16) & 0xFF);
            out.write((pack >>> 8)  & 0xFF);
            out.write((pack >>> 0)  & 0xFF);
            bufPos = 0;
        }
    }
}
	

First, CompressionOutputStream subclasses FilterOutputStream. Then it implements the interface CompressionConstants so that it can have access to the lookup table and constants.

In order to compress the data, the FilterOutputStream write methods are overridden by CompressionOutputStream. The method

public void write(int b)

writes one character per invocation, and is responsible for marking each character with its encoding format using a compression constant (either RAW or BASE) and dividing the character into two 4-bit parts when necessary.

The method

public void write(byte b[], int off, int len)

allows one to write len characters. It calls the write method that writes one character per invocation len times.

The method writeCode is responsible for packing five 6-bit codes into one word (which could encode up to five characters) and writing that word to the output stream.

The source code for CompressionOutputStream can be found here.

Now that we have an output stream that compresses the data, we need to implement an input stream that decompresses it. Creating the CompressionInputStream is very similar to creating the CompressionOutputStream as you will notice from the source code and the following discussion.

And, of course, algorithmically, the decoding process is the just reverse of the encoding process.

import java.io.*

class CompressionInputStream extends FilterInputStream
    implements CompressionConstants
{
    /*
     * Constructor calls constructor of superclass
     */
    public CompressionInputStream(InputStream in) {
        super(in);
    }
 
    /* 
     * Buffer of unpacked 6-bit codes 
     * from last 32 bits read.
     */
    int buf[] = new int[5];
 
    /*
     * Position of next code to read in buffer (5 signifies end). 
     */ 
    int bufPos = 5;
 
    /*
     * Reads in format code and decompresses character
     * accordingly.  
     */
    public int read() throws IOException {
        try {
            int code;

            // Read in and ignore empty bytes (NOP's) as 
            // long as they arrive. 
            do {
                code = readCode();
            } while (code == NOP);      
 
            if (code >= BASE) {
                // Retrieve index of character in codeTable 
                // if the code is in the correct range.

                return codeTable.charAt(code - BASE);
            } else if (code == RAW) {
                // read in the lower 4 bits and the 
                // higher 4 bits, and return the 
                // reconstructed character
                int high = readCode();
                int low = readCode();
                return (high << 4) | low;
            } else 
                throw new IOException("unknown compression code: " 
                                      + code);
        } catch (EOFException e) {
            // Return the end of file code
            return -1;
        }
    }
 
    /* 
     * This method reads up to len bytes from the input stream.
     * Returns if read blocks before len bytes are read. 
     */ 
    public int read(byte b[], int off, int len) 
        throws IOException 
    {
        if (len <= 0) {
            return 0;
        }
 
        // Read in a word and return -1 if no more data.
        int c = read();
        if (c == -1) {
            return -1;
        }

        // Save c in buffer b
        b[off] = (byte)c;
 
        int i = 1;
        // Try to read up to len bytes or until no
        // more bytes can be read without blocking.
        try {
            for (; (i < len) && (in.available() > 0); i++) {
                c = read();
                if (c == -1) {
                    break;
                }
                if (b != null) {
                    b[off + i] = (byte)c;
                }
            }
        } catch (IOException ee) {
        }
        return i;
    }

    /*
     * If there is no more data to decode left
     * in buf, read the next four bytes from the 
     * wire. Then store each group of 6 bits in an
     * element of buf.  Return one element of buf.
     */
    private int readCode() throws IOException {
        // As soon as all the data in buf has been read
        // (when bufPos == 5) read in another four bytes.
        if (bufPos == 5) {
            int b1 = in.read();
            int b2 = in.read();
            int b3 = in.read();
            int b4 = in.read();

            // make sure none of the bytes signify the
            // end of the data in the stream
            if ((b1 | b2 | b3 | b4) < 0)
                throw new EOFException();
            // Assign each group of 6 bits to an 
            // element of buf.
            int pack = (b1 << 24) | (b2 << 16) | 
                       (b3 << 8) | b4;
            buf[0] = (pack >>> 24) & 0x3F;
            buf[1] = (pack >>> 18) & 0x3F;
            buf[2] = (pack >>> 12) & 0x3F;
            buf[3] = (pack >>>  6) & 0x3F;
            buf[4] = (pack >>>  0) & 0x3F;
            bufPos = 0;
        }
        return buf[bufPos++];
    }
}

When writing your own input stream it is necessary to provide methods for getting data from the stream. Therefore, in addition to a constructor, two FilterOutputStream read methods were overridden. The method

public int read()

reads one character per invocation and is responsible for decoding the 6-bit codes unpacked by the method writeCode.

The method

public int read(byte b[], int off, int len)

causes up to len bytes to be read and placed into the array b. It accomplishes this through up to len calls to the read method that reads one character per invocation. Besides returning once len bytes have been read, or once the end of the file has been reached, this method also returns as soon as no more bytes can be read without blocking.

The method readcode, which corresponds to the writecode method in CompressionOutputStream, is responsible for reading data from the stream and unpacking each group of four bytes into five 6-bit codes (which are then decoded by read).

The source code for CompressionInputStream.java, can be found here.

Now that we have implemented the classes CompressionInputStream and CompressionOutputStream, we can implement the socket that communicates using these compression streams. Our subclass extends class java.net.Socket.

Below is the source code for the class CompressionSocket. Following the source is a discussion of the class.

import java.io.*;
import java.net.*;
 
class CompressionSocket extends Socket {

    /* InputStream used by socket */
    private InputStream in;
    /* OutputStream used by socket */
    private OutputStream out;

    /* 
     * No-arg constructor for class CompressionSocket  
     */
    public CompressionSocket() { super(); }

    /* 
     * Constructor for class CompressionSocket 
     */
    public CompressionSocket(String host, int port) 
        throws IOException 
    {
        super(host, port);
    }

    /* 
     * Returns a stream of type CompressionInputStream 
     */
    public InputStream getInputStream() 
        throws IOException 
    {
        if (in == null) {
            in = new CompressionInputStream(super.getInputStream());
        }
        return in;
    }

    /* 
     * Returns a stream of type CompressionOutputStream 
     */
    public OutputStream getOutputStream() 
        throws IOException 
    {
        if (out == null) {
            out = new CompressionOutputStream(super.getOutputStream());
        }
        return out;
    }

    /*
     * Flush the CompressionOutputStream before 
     * closing the socket.
     */
    public synchronized void close() throws IOException {
        OutputStream o = getOutputStream();
        o.flush();
	super.close();
    }

}
      

Since we are extending the Socket class to provide sockets that communicate using data compression, we need to override the class Socket methods that directly manipulate the input and output streams used by class Socket and provide constructors that call the constructor of the super class.

The CompressionSocket constructors just call the equivalent constructors in the super class, java.net.Socket.

The method getInputStream creates a CompressionInputStream for the socket if one has not already been instantiated and returns a reference to that stream. Likewise, getOutputStream creates a CompressionOutputStream, if necessary, and returns the one in use by the CompressionSocket.

The close method flushes the underlying CompressionOutputStream, ensuring that all the data is sent before the socket is closed.

The source code for CompressionSocket.java, can be found here.

The last step in writing a custom socket is creating a subclass of ServerSocket that supports your protocol. In this case, our subclass will be CompressionServerSocket.

Below is the source code for the CompressionServerSocket class, followed by a discussion of the class.

import java.io.*;
import java.net.*;
 
class CompressionServerSocket extends ServerSocket {

    public CompressionServerSocket(int port) 
        throws IOException 
    {
        super(port);
    }

    public Socket accept() throws IOException {
        Socket s = new CompressionSocket();
        implAccept(s);
        return s;
    }
}

As was the case with CompressionSocket, in order to provide server sockets that communicate using our compression protocol, we need to implement the constructor, and then make sure that all methods that use sockets of type java.net.Socket are overridden to use sockets of type CompressionSocket.

Implementing the constructor is as simple as calling the constructor of the super class.

The only class ServerSocket method that needs to be overridden is the method accept. It is overridden to instantiate a socket of type CompressionSocket instead of type Socket.

The above simplifications are possible because the compression socket type described in this tutorial is a protocol layer on top of TCP, the default protocol used by java.net.Socket and java.net.ServerSocket. Accordingly, it shares the same meaning for the rest of the methods, as well as a similar connection-establishment interface.

The source code for CompressionServerSocket.java, can be found here.