Most physical networks are inherently unreliable. They can drop data, corrupt it, deliver it out of order, and deliver multiple copies of it. Some transport protocols hide this unreliability from applications by using protocol-level mechanisms to ensure that data is delivered in the correct order, without loss, corruption, or duplication. Other transport protocols do not make these guarantees; they simply make a best effort to get the data to its intended recipient, but physical network failures can lead to data integrity problems.
Because they have less inherent overhead, unreliable protocols often provide better performance, especially for applications which send small amounts of data in a request-response (transaction) format. However, connectionless protocols impose the burden on the application of doing whatever reliability guarantees are necessary, and therefore lead to more complicated applications. Reliable protocols are usually simpler for the application at the expense of a higher cost in system resources and performance.
Typically, connection-oriented protocols are reliable, while connectionless protocols are unreliable. While this is true of all the transport protocols supported by Windows NT, it is not required that these always go hand-in-hand.