Blocking vs. Non-Blocking Sockets

One of the first issues that you’ll encounter when developing your Windows Sockets applications is the difference between blocking and non-blocking sockets. Whenever you perform some operation on a socket, it may not be able to complete immediately and return control back to your program. For example, a read on a socket cannot complete until some data has been sent by the remote host. If there is no data waiting to be read, one of two things can happen: the function can wait until some data has been written on the socket, or it can return immediately with an error that indicates that there is no data to be read.

The first case is called a blocking socket. In other words, the program is "blocked" until the request for data has been satisfied. When the remote system does write some data on the socket, the read operation will complete and execution of the program will resume. The second case is called a non-blocking socket, and requires that the application recognize the error condition and handle the situation appropriately. Programs that use non-blocking sockets typically use one of two methods when sending and receiving data. The first method, called polling, is when the program periodically attempts to read or write data from the socket (typically using a timer). The second, and preferred method, is to use what is called asynchronous notification. This means that the program is notified whenever a socket event takes place, and in turn can respond to that event. For example, if the remote program writes some data to the socket, a "read event" is generated so that program knows it can read the data from the socket at that point.

For historical reasons, the default behavior is for socket functions to "block" and not return until the operation has completed. However, blocking sockets in Windows can introduce some special problems. For 16-bit applications, the blocking function will enter what is called a "message loop" where it continues to process messages sent to it by Windows and other applications. Since messages are being processed, this means that the program can be re-entered at a different point with the blocked operation parked on the program's stack. For example, consider a program that attempts to read some data from the socket when a button is pressed. Because no data has been written yet, it blocks and the program goes into a message loop. The user then presses a different button, which causes code to be executed, which in turn attempts to read data from the socket, and so on.

Blocking socket functions can introduce a different type of problem in 32-bit applications because blocking functions will prevent the calling thread from processing any messages sent to it. Since many applications are single-threaded, this can result in the application being unresponsive to user actions. To resolve the general problems with blocking sockets, the Windows Sockets standard states that there may only be one outstanding blocked call per thread of execution. This means that 16-bit applications that are re-entered (as in the example above) will encounter errors whenever they try to take some action while a blocking function is already in progress. With 32-bit programs, the creation of worker threads to perform blocking socket operations is a common approach, although it introduces additional complexity into the application.

The SocketWrench control facilitates the use of non-blocking sockets by firing events. For example, a Read event is generated whenever the remote host writes on the socket, which tells your application that there is data waiting to be read. The use of non-blocking sockets will be demonstrated in the next section, and is one of the key areas in which a control has a distinct advantage over coding directly against the Windows Sockets API.

In summary, there are three general approaches that can be taken when building an application with the control in regard to blocking or non-blocking sockets:

• Use a blocking (synchronous) socket. In this mode, the program will not resume execution until the socket operation has completed. Blocking sockets in 16-bit application will allow it to be re-entered at a different point, and 32-bit applications will stop responding to user actions. This can lead to complex interactions (and difficult debugging) if there are multiple active controls in use by the application.

• Use a non-blocking (asynchronous) socket, which allows your application to respond to events. For example, when the remote system writes data to the socket, a Read event is generated for the control. Your application can respond by reading the data from the socket, and perhaps send some data back, depending on the context of the data received.

• Use a combination of blocking and non-blocking socket operations. The ability to switch between blocking and non-blocking modes "on the fly" provides a powerful and convenient way to perform socket operations. Note that the warning regarding blocking sockets also applies here.

If you decide to use non-blocking sockets in your application, it’s important to keep in mind that you must check the return value from every read and write operation, since it is possible that you may not be able to send or receive all of the specified data. Frequently, developers encounter problems when they write a program that assumes a given number of bytes can always be written to, or read from, the socket. In many cases, the program works as expected when developed and tested on a local area network, but fails unpredictably when the program is released to a user that has a slower network connection (such as a serial dial-up connection to the Internet). By always checking the return values of these operations, you insure that your program will work correctly, regardless of the speed or configuration of the network.