Nagle's algorithm

Nagle's algorithm is a method for improving the efficiency of TCP/IP networks by reducing how many small packets are sent. It was defined by John Nagle while he worked at Ford Aerospace and was published in 1984 as RFC 896, titled Congestion Control in IP/TCP Internetworks. The RFC describes what Nagle called the small-packet problem: an application that repeatedly emits data in tiny chunks, often a single byte, produces a 41-byte packet for one byte of useful data, because the TCP and IPv4 headers take 40 bytes. The situation was common in Telnet sessions, and over slow links many such packets in transit could lead to congestion collapse. The algorithm combines small outgoing messages and sends them together: while a sent packet remains unacknowledged, the sender keeps buffering its output until it has a full packet's worth. It interacts badly with TCP delayed acknowledgments, which can add a constant delay of up to 500 milliseconds when an application makes two writes followed by a read, so applications that need low latency, such as multiplayer games, usually disable it with a socket option that switches the algorithm off. Minshall's modification addresses the large-write case.