Socket选项

对一个的java类为java.net.SocketOptions

TCP_NODELAY

Fetch the local address binding of a socket (this option cannot be “set” only “gotten”, since sockets are bound at creation time, and so the locally bound address cannot be changed). The default local address of a socket is INADDR_ANY, meaning any local address on a multi-homed host. A multi-homed host can use this option to accept connections to only one of its addresses (in the case of a ServerSocket or DatagramSocket), or to specify its return address to the peer (for a Socket or DatagramSocket). The parameter of this option is an InetAddress.
This option must be specified in the constructor.

/**
 * Disable Nagle's algorithm for this connection.  Written data
 * to the network is not buffered pending acknowledgement of
 * previously written data.
 *<P>
 * Valid for TCP only: SocketImpl.
 *
 * @see Socket#setTcpNoDelay
 * @see Socket#getTcpNoDelay
 */
@Native public final static int TCP_NODELAY = 0x0001;

设置TCP_NODELAY为true可确保包无论大小会尽可能快地发送。正常情况下，小数据包在发送前会组合为更大的包。另外，在发送另一个包之前，本地主机需要等待远程系统对前一个包的确认，这称为Nagle算法。Nagle算法的问题是，如果远程系统没有足够快的将确认发送会本地系统，那么依赖于小数据量信息稳定传输的应用程序会变得很慢，对于GUI程序，比如游戏或网络极端及应用程序（服务器需要实时跟踪客户端鼠标的移动）。这个问题尤其严重，在一个相当慢的网络中，即使简单地打字也会由于持续的缓冲而变得太慢。设置TCP_NODELAY为true可以打破这种缓冲模式，这样所有包一旦就绪就会发送。

SO_LINGER

/**
 * Specify a linger-on-close timeout.  This option disables/enables
 * immediate return from a <B>close()</B> of a TCP Socket.  Enabling
 * this option with a non-zero Integer <I>timeout</I> means that a
 * <B>close()</B> will block pending the transmission and acknowledgement
 * of all data written to the peer, at which point the socket is closed
 * <I>gracefully</I>.  Upon reaching the linger timeout, the socket is
 * closed <I>forcefully</I>, with a TCP RST. Enabling the option with a
 * timeout of zero does a forceful close immediately. If the specified
 * timeout value exceeds 65,535 it will be reduced to 65,535.
 * <P>
 * Valid only for TCP: SocketImpl
 *
 * @see Socket#setSoLinger
 * @see Socket#getSoLinger
 */
@Native public final static int SO_LINGER = 0x0080;

SO_LINGER选项指定了Socket关闭时如何处理尚未发送的数据报。默认情况下，close方法将立刻返回，但系统仍会尝试发送剩余的数据，如果延迟时间设置为0，那么当Socket关闭时，所有未发送的数据包都将被丢弃。如果SO_LINGER打开而且延迟时间设置为任意整数，close方法会阻塞（阻塞时间为指定的秒数），等待发送数据和接收确认。当过去相应秒数后，Socket关闭，所有剩余的数据都不会发送，也不会收到确认。

如果底层Socket实现不支持SO_LINGER选项，setSoLinger和getSoLinger这两个方法都会抛出SocketException异常，如果试图将延迟时间设置为一个负数，setSoLinger方法还会抛出一个IllegalArgumentException异常。

/**
 * Enable/disable {@link SocketOptions#SO_LINGER SO_LINGER} with the
 * specified linger time in seconds. The maximum timeout value is platform
 * specific.
 *
 * The setting only affects socket close.
 *
 * @param on     whether or not to linger on.
 * @param linger how long to linger for, if on is true.
 * @exception SocketException if there is an error
 * in the underlying protocol, such as a TCP error.
 * @exception IllegalArgumentException if the linger value is negative.
 * @since JDK1.1
 * @see #getSoLinger()
 */
public void setSoLinger(boolean on, int linger) throws SocketException {
    if (isClosed())
        throw new SocketException("Socket is closed");
    if (!on) {
        getImpl().setOption(SocketOptions.SO_LINGER, new Boolean(on));
    } else {
        if (linger < 0) {
            throw new IllegalArgumentException("invalid value for SO_LINGER");
        }
        if (linger > 65535)
            linger = 65535;
        getImpl().setOption(SocketOptions.SO_LINGER, new Integer(linger));
    }
}

不过，getSoLinger方法可以返回-1，指示这个选项被禁用，会根据需要用更多时间发送剩余的数据。

/**
 * Returns setting for {@link SocketOptions#SO_LINGER SO_LINGER}.
 * -1 returns implies that the
 * option is disabled.
 *
 * The setting only affects socket close.
 *
 * @return the setting for {@link SocketOptions#SO_LINGER SO_LINGER}.
 * @exception SocketException if there is an error
 * in the underlying protocol, such as a TCP error.
 * @since   JDK1.1
 * @see #setSoLinger(boolean, int)
 */
public int getSoLinger() throws SocketException {
    if (isClosed())
        throw new SocketException("Socket is closed");
    Object o = getImpl().getOption(SocketOptions.SO_LINGER);
    if (o instanceof Integer) {
        return ((Integer) o).intValue();
    } else {
        return -1;
    }
}

SO_RCVBUF和SO_SNDBUF

TCP使用缓冲区提升网络性能。较大的缓冲区会提升较高带宽（比如10Mb/s）的性能。而较慢的带宽利用较少的缓冲区会有更好的表现。一般来讲，传输连续的大数据块时（这在文件传输协议如FTP和HTTP中很常见），可以从大缓冲区受益；而对于交互式会话的小数据量传输（如Telnet和很多游戏），大缓冲区则没有多大帮助。一些相对较老的操作系统在小文件和慢速网络时代设计的，如BSD 4.2使用2KB字节的缓冲区，Windows XP使用17 520字节的缓冲区。如今，128KB字节已经是一个常见的默认设置。

可以达到的最大带宽等于缓冲区大小除以延迟。例如，在Windows XP上，假设两个主机之间的延迟为半秒（500ms），则带宽为17520字节/0.5秒=35040字节/秒=273.75kb/s。这就是Socket的最大速度。而不论网络速度有多块，这对于一个拨号连接来说相当快，对于ISDN也不错，不过对于DSL或FIOS就不够了。

可以通过减少延迟来提高速度，不够，延迟与网络硬件有关。另外还取决于你的应用控制之外的其他一些因素。另一方面，缓冲区的大小是可以控制的，例如，如果把缓冲区的大小从17520字节增加到128字节，最大带宽就会增加到2Mb/s。将缓冲区大小再次增加到256KB字节，那么最大带宽就会加倍到4Mb/s。当然，网络本身对最大带宽也有限制的，将缓冲区大小设置得过高，程序会试图以过高的速度发送和接收数据，而网络来不及处理，这就会导致拥塞、丢包和性能下降。因此，要得到最大带宽，需要让缓冲区大小与连接的延迟匹配，使她稍小于网络的带宽。

可以使用ping手动检查某个特定主机的延迟，或者可以在程序中测量InetAddress.isReachable调用的时间。

/**
 * Set a hint the size of the underlying buffers used by the
 * platform for outgoing network I/O. When used in set, this is a
 * suggestion to the kernel from the application about the size of
 * buffers to use for the data to be sent over the socket. When
 * used in get, this must return the size of the buffer actually
 * used by the platform when sending out data on this socket.
 *
 * Valid for all sockets: SocketImpl, DatagramSocketImpl
 *
 * @see Socket#setSendBufferSize
 * @see Socket#getSendBufferSize
 * @see DatagramSocket#setSendBufferSize
 * @see DatagramSocket#getSendBufferSize
 */
@Native public final static int SO_SNDBUF = 0x1001;

/**
 * Set a hint the size of the underlying buffers used by the
 * platform for incoming network I/O. When used in set, this is a
 * suggestion to the kernel from the application about the size of
 * buffers to use for the data to be received over the
 * socket. When used in get, this must return the size of the
 * buffer actually used by the platform when receiving in data on
 * this socket.
 *
 * Valid for all sockets: SocketImpl, DatagramSocketImpl
 *
 * @see Socket#setReceiveBufferSize
 * @see Socket#getReceiveBufferSize
 * @see DatagramSocket#setReceiveBufferSize
 * @see DatagramSocket#getReceiveBufferSize
 */
@Native public final static int SO_RCVBUF = 0x1002;

SO_RCVBUF选项控制用于网络输入的建议的接收缓冲区大小。SO_SNDBUF选项控制用于网络输入的建议的发送缓冲区大小。尽管看起来可以单独设置发送和缓冲区大小，但实际上缓冲区通常会设置为二者中较小的一个，例如，如果将发送缓冲区设置为64KB，而接收缓冲区设置为128KB，那么发送和接收缓冲区的大小都将是64KB。Java的相关方法会返回接收缓冲区为128KB，但底层TCP栈实际上会使用64KB。

setReceiveBufferSize或者setSendBufferSize方法会对Socket上缓冲输出使用的字节数给出一个建议，不过，底层实现完全可以忽略或调整这个建议。具体的，UNIX和Linux系统通常指定一个最大缓冲区大小，一般是64KB或256KB，而且不允许任何Socket有更大的缓冲区。如果你试图设置一个更大的值，Java会把它调整为可能的最大缓冲区大小。在Linux上，底层实现由可能将请求的缓冲区大小加倍，这也并非没有先例。例如，如果你请求一个64KB缓冲区，可能会得到一个128KB的缓冲区。

一般来讲，如果你发现你的应用不能充分利用可用带宽（例如，你有一个25Mb/sd的网络连接，但是数据传输率仅仅为1.5Mb/s）,那么可以试着增加缓冲区大小。相反，如果存在丢包和拥塞现象，则需要减少缓冲区大小。不过，大多数情况下， 除非网络在某个方向上负载过大， 否则默认值就很合适。具体来说，现在操作系统使用TCP滑动窗口来动态调整缓冲区大小，以适应网络。与几乎所有性能调优建议一样，一般经验是除非你检测到了某个问题，否则不要进行调整。即使非要调整，可以在操作系统级增加允许的最大缓冲区大小，与调整单个socket的缓冲区大小相比，前者可以得到更大的速度提升。

java.net.StandardSocketOptions

package java.net;

/**
 * Defines the <em>standard</em> socket options.
 *
 * <p> The {@link SocketOption#name name} of each socket option defined by this
 * class is its field name.
 *
 * <p> In this release, the socket options defined here are used by {@link
 * java.nio.channels.NetworkChannel network} channels in the {@link
 * java.nio.channels channels} package.
 *
 * @since 1.7
 */

public final class StandardSocketOptions {
    private StandardSocketOptions() { }

    // -- SOL_SOCKET --

    /**
     * Allow transmission of broadcast datagrams.
     *
     * <p> The value of this socket option is a {@code Boolean} that represents
     * whether the option is enabled or disabled. The option is specific to
     * datagram-oriented sockets sending to {@link java.net.Inet4Address IPv4}
     * broadcast addresses. When the socket option is enabled then the socket
     * can be used to send <em>broadcast datagrams</em>.
     *
     * <p> The initial value of this socket option is {@code FALSE}. The socket
     * option may be enabled or disabled at any time. Some operating systems may
     * require that the Java virtual machine be started with implementation
     * specific privileges to enable this option or send broadcast datagrams.
     *
     * @see <a href="http://www.ietf.org/rfc/rfc919.txt">RFC&nbsp;929:
     * Broadcasting Internet Datagrams</a>
     * @see DatagramSocket#setBroadcast
     */
    public static final SocketOption<Boolean> SO_BROADCAST =
        new StdSocketOption<Boolean>("SO_BROADCAST", Boolean.class);

    /**
     * Keep connection alive.
     *
     * <p> The value of this socket option is a {@code Boolean} that represents
     * whether the option is enabled or disabled. When the {@code SO_KEEPALIVE}
     * option is enabled the operating system may use a <em>keep-alive</em>
     * mechanism to periodically probe the other end of a connection when the
     * connection is otherwise idle. The exact semantics of the keep alive
     * mechanism is system dependent and therefore unspecified.
     *
     * <p> The initial value of this socket option is {@code FALSE}. The socket
     * option may be enabled or disabled at any time.
     *
     * @see <a href="http://www.ietf.org/rfc/rfc1122.txt">RFC&nbsp;1122
     * Requirements for Internet Hosts -- Communication Layers</a>
     * @see Socket#setKeepAlive
     */
    public static final SocketOption<Boolean> SO_KEEPALIVE =
        new StdSocketOption<Boolean>("SO_KEEPALIVE", Boolean.class);

    /**
     * The size of the socket send buffer.
     *
     * <p> The value of this socket option is an {@code Integer} that is the
     * size of the socket send buffer in bytes. The socket send buffer is an
     * output buffer used by the networking implementation. It may need to be
     * increased for high-volume connections. The value of the socket option is
     * a <em>hint</em> to the implementation to size the buffer and the actual
     * size may differ. The socket option can be queried to retrieve the actual
     * size.
     *
     * <p> For datagram-oriented sockets, the size of the send buffer may limit
     * the size of the datagrams that may be sent by the socket. Whether
     * datagrams larger than the buffer size are sent or discarded is system
     * dependent.
     *
     * <p> The initial/default size of the socket send buffer and the range of
     * allowable values is system dependent although a negative size is not
     * allowed. An attempt to set the socket send buffer to larger than its
     * maximum size causes it to be set to its maximum size.
     *
     * <p> An implementation allows this socket option to be set before the
     * socket is bound or connected. Whether an implementation allows the
     * socket send buffer to be changed after the socket is bound is system
     * dependent.
     *
     * @see Socket#setSendBufferSize
     */
    public static final SocketOption<Integer> SO_SNDBUF =
        new StdSocketOption<Integer>("SO_SNDBUF", Integer.class);


    /**
     * The size of the socket receive buffer.
     *
     * <p> The value of this socket option is an {@code Integer} that is the
     * size of the socket receive buffer in bytes. The socket receive buffer is
     * an input buffer used by the networking implementation. It may need to be
     * increased for high-volume connections or decreased to limit the possible
     * backlog of incoming data. The value of the socket option is a
     * <em>hint</em> to the implementation to size the buffer and the actual
     * size may differ.
     *
     * <p> For datagram-oriented sockets, the size of the receive buffer may
     * limit the size of the datagrams that can be received. Whether datagrams
     * larger than the buffer size can be received is system dependent.
     * Increasing the socket receive buffer may be important for cases where
     * datagrams arrive in bursts faster than they can be processed.
     *
     * <p> In the case of stream-oriented sockets and the TCP/IP protocol, the
     * size of the socket receive buffer may be used when advertising the size
     * of the TCP receive window to the remote peer.
     *
     * <p> The initial/default size of the socket receive buffer and the range
     * of allowable values is system dependent although a negative size is not
     * allowed. An attempt to set the socket receive buffer to larger than its
     * maximum size causes it to be set to its maximum size.
     *
     * <p> An implementation allows this socket option to be set before the
     * socket is bound or connected. Whether an implementation allows the
     * socket receive buffer to be changed after the socket is bound is system
     * dependent.
     *
     * @see <a href="http://www.ietf.org/rfc/rfc1323.txt">RFC&nbsp;1323: TCP
     * Extensions for High Performance</a>
     * @see Socket#setReceiveBufferSize
     * @see ServerSocket#setReceiveBufferSize
     */
    public static final SocketOption<Integer> SO_RCVBUF =
        new StdSocketOption<Integer>("SO_RCVBUF", Integer.class);

    /**
     * Re-use address.
     *
     * <p> The value of this socket option is a {@code Boolean} that represents
     * whether the option is enabled or disabled. The exact semantics of this
     * socket option are socket type and system dependent.
     *
     * <p> In the case of stream-oriented sockets, this socket option will
     * usually determine whether the socket can be bound to a socket address
     * when a previous connection involving that socket address is in the
     * <em>TIME_WAIT</em> state. On implementations where the semantics differ,
     * and the socket option is not required to be enabled in order to bind the
     * socket when a previous connection is in this state, then the
     * implementation may choose to ignore this option.
     *
     * <p> For datagram-oriented sockets the socket option is used to allow
     * multiple programs bind to the same address. This option should be enabled
     * when the socket is to be used for Internet Protocol (IP) multicasting.
     *
     * <p> An implementation allows this socket option to be set before the
     * socket is bound or connected. Changing the value of this socket option
     * after the socket is bound has no effect. The default value of this
     * socket option is system dependent.
     *
     * @see <a href="http://www.ietf.org/rfc/rfc793.txt">RFC&nbsp;793: Transmission
     * Control Protocol</a>
     * @see ServerSocket#setReuseAddress
     */
    public static final SocketOption<Boolean> SO_REUSEADDR =
        new StdSocketOption<Boolean>("SO_REUSEADDR", Boolean.class);

    /**
     * Linger on close if data is present.
     *
     * <p> The value of this socket option is an {@code Integer} that controls
     * the action taken when unsent data is queued on the socket and a method
     * to close the socket is invoked. If the value of the socket option is zero
     * or greater, then it represents a timeout value, in seconds, known as the
     * <em>linger interval</em>. The linger interval is the timeout for the
     * {@code close} method to block while the operating system attempts to
     * transmit the unsent data or it decides that it is unable to transmit the
     * data. If the value of the socket option is less than zero then the option
     * is disabled. In that case the {@code close} method does not wait until
     * unsent data is transmitted; if possible the operating system will transmit
     * any unsent data before the connection is closed.
     *
     * <p> This socket option is intended for use with sockets that are configured
     * in {@link java.nio.channels.SelectableChannel#isBlocking() blocking} mode
     * only. The behavior of the {@code close} method when this option is
     * enabled on a non-blocking socket is not defined.
     *
     * <p> The initial value of this socket option is a negative value, meaning
     * that the option is disabled. The option may be enabled, or the linger
     * interval changed, at any time. The maximum value of the linger interval
     * is system dependent. Setting the linger interval to a value that is
     * greater than its maximum value causes the linger interval to be set to
     * its maximum value.
     *
     * @see Socket#setSoLinger
     */
    public static final SocketOption<Integer> SO_LINGER =
        new StdSocketOption<Integer>("SO_LINGER", Integer.class);


    // -- IPPROTO_IP --

    /**
     * The Type of Service (ToS) octet in the Internet Protocol (IP) header.
     *
     * <p> The value of this socket option is an {@code Integer} representing
     * the value of the ToS octet in IP packets sent by sockets to an {@link
     * StandardProtocolFamily#INET IPv4} socket. The interpretation of the ToS
     * octet is network specific and is not defined by this class. Further
     * information on the ToS octet can be found in <a
     * href="http://www.ietf.org/rfc/rfc1349.txt">RFC&nbsp;1349</a> and <a
     * href="http://www.ietf.org/rfc/rfc2474.txt">RFC&nbsp;2474</a>. The value
     * of the socket option is a <em>hint</em>. An implementation may ignore the
     * value, or ignore specific values.
     *
     * <p> The initial/default value of the TOS field in the ToS octet is
     * implementation specific but will typically be {@code 0}. For
     * datagram-oriented sockets the option may be configured at any time after
     * the socket has been bound. The new value of the octet is used when sending
     * subsequent datagrams. It is system dependent whether this option can be
     * queried or changed prior to binding the socket.
     *
     * <p> The behavior of this socket option on a stream-oriented socket, or an
     * {@link StandardProtocolFamily#INET6 IPv6} socket, is not defined in this
     * release.
     *
     * @see DatagramSocket#setTrafficClass
     */
    public static final SocketOption<Integer> IP_TOS =
        new StdSocketOption<Integer>("IP_TOS", Integer.class);

    /**
     * The network interface for Internet Protocol (IP) multicast datagrams.
     *
     * <p> The value of this socket option is a {@link NetworkInterface} that
     * represents the outgoing interface for multicast datagrams sent by the
     * datagram-oriented socket. For {@link StandardProtocolFamily#INET6 IPv6}
     * sockets then it is system dependent whether setting this option also
     * sets the outgoing interface for multicast datagrams sent to IPv4
     * addresses.
     *
     * <p> The initial/default value of this socket option may be {@code null}
     * to indicate that outgoing interface will be selected by the operating
     * system, typically based on the network routing tables. An implementation
     * allows this socket option to be set after the socket is bound. Whether
     * the socket option can be queried or changed prior to binding the socket
     * is system dependent.
     *
     * @see java.nio.channels.MulticastChannel
     * @see MulticastSocket#setInterface
     */
    public static final SocketOption<NetworkInterface> IP_MULTICAST_IF =
        new StdSocketOption<NetworkInterface>("IP_MULTICAST_IF", NetworkInterface.class);

    /**
     * The <em>time-to-live</em> for Internet Protocol (IP) multicast datagrams.
     *
     * <p> The value of this socket option is an {@code Integer} in the range
     * {@code 0 <= value <= 255}. It is used to control the scope of multicast
     * datagrams sent by the datagram-oriented socket.
     * In the case of an {@link StandardProtocolFamily#INET IPv4} socket
     * the option is the time-to-live (TTL) on multicast datagrams sent by the
     * socket. Datagrams with a TTL of zero are not transmitted on the network
     * but may be delivered locally. In the case of an {@link
     * StandardProtocolFamily#INET6 IPv6} socket the option is the
     * <em>hop limit</em> which is number of <em>hops</em> that the datagram can
     * pass through before expiring on the network. For IPv6 sockets it is
     * system dependent whether the option also sets the <em>time-to-live</em>
     * on multicast datagrams sent to IPv4 addresses.
     *
     * <p> The initial/default value of the time-to-live setting is typically
     * {@code 1}. An implementation allows this socket option to be set after
     * the socket is bound. Whether the socket option can be queried or changed
     * prior to binding the socket is system dependent.
     *
     * @see java.nio.channels.MulticastChannel
     * @see MulticastSocket#setTimeToLive
     */
    public static final SocketOption<Integer> IP_MULTICAST_TTL =
        new StdSocketOption<Integer>("IP_MULTICAST_TTL", Integer.class);

    /**
     * Loopback for Internet Protocol (IP) multicast datagrams.
     *
     * <p> The value of this socket option is a {@code Boolean} that controls
     * the <em>loopback</em> of multicast datagrams. The value of the socket
     * option represents if the option is enabled or disabled.
     *
     * <p> The exact semantics of this socket options are system dependent.
     * In particular, it is system dependent whether the loopback applies to
     * multicast datagrams sent from the socket or received by the socket.
     * For {@link StandardProtocolFamily#INET6 IPv6} sockets then it is
     * system dependent whether the option also applies to multicast datagrams
     * sent to IPv4 addresses.
     *
     * <p> The initial/default value of this socket option is {@code TRUE}. An
     * implementation allows this socket option to be set after the socket is
     * bound. Whether the socket option can be queried or changed prior to
     * binding the socket is system dependent.
     *
     * @see java.nio.channels.MulticastChannel
     *  @see MulticastSocket#setLoopbackMode
     */
    public static final SocketOption<Boolean> IP_MULTICAST_LOOP =
        new StdSocketOption<Boolean>("IP_MULTICAST_LOOP", Boolean.class);


    // -- IPPROTO_TCP --

    /**
     * Disable the Nagle algorithm.
     *
     * <p> The value of this socket option is a {@code Boolean} that represents
     * whether the option is enabled or disabled. The socket option is specific to
     * stream-oriented sockets using the TCP/IP protocol. TCP/IP uses an algorithm
     * known as <em>The Nagle Algorithm</em> to coalesce short segments and
     * improve network efficiency.
     *
     * <p> The default value of this socket option is {@code FALSE}. The
     * socket option should only be enabled in cases where it is known that the
     * coalescing impacts performance. The socket option may be enabled at any
     * time. In other words, the Nagle Algorithm can be disabled. Once the option
     * is enabled, it is system dependent whether it can be subsequently
     * disabled. If it cannot, then invoking the {@code setOption} method to
     * disable the option has no effect.
     *
     * @see <a href="http://www.ietf.org/rfc/rfc1122.txt">RFC&nbsp;1122:
     * Requirements for Internet Hosts -- Communication Layers</a>
     * @see Socket#setTcpNoDelay
     */
    public static final SocketOption<Boolean> TCP_NODELAY =
        new StdSocketOption<Boolean>("TCP_NODELAY", Boolean.class);


    private static class StdSocketOption<T> implements SocketOption<T> {
        private final String name;
        private final Class<T> type;
        StdSocketOption(String name, Class<T> type) {
            this.name = name;
            this.type = type;
        }
        @Override public String name() { return name; }
        @Override public Class<T> type() { return type; }
        @Override public String toString() { return name; }
    }
}