Connection-oriented protocol

From Wikipedia, the free encyclopedia

A connection-oriented networking protocol is one which identifies traffic flows by some connection identifier rather than by explicitly listing source and destination addresses. Typically, this connection identifier is a small integer (10 bits for Frame Relay, 24 for ATM, for example). This makes network switches substantially faster (as routing tables are just simple look-up tables, and are trivial to implement in hardware). The impact is so great, in fact, that even characteristically connectionless protocols, such as IP traffic, are being tagged with connection-oriented header prefixes (e.g., as with MPLS, or IPv6's built-in Flow ID field).

Note that connection-oriented protocols are not necessarily reliable protocols. ATM and Frame Relay, for example, are both examples of a connection-oriented, unreliable protocol. There are also reliable connectionless protocols as well, such as AX.25 when it passes data in I-frames. But this combination is rare, and reliable-connectionless is uncommon in commercial and academic networks.

Note that connection-oriented protocols handle real-time traffic substantially more efficiently than connectionless protocols, which is why ATM has yet to be replaced by Ethernet for carrying real-time, isochronous traffic streams, especially in heavily aggregated networks like backbones, where the motto "bandwidth is cheap" fails to deliver on its promise. Experience has also shown that overprovisioning bandwidth does not resolve all quality of service issues. Hence, (10-)gigabit Ethernet is not expected to replace ATM at this time.

[edit] See also