John Nagle. Rfc896: Congestion control in ip/tcp internetworks. Technical report, Internet Assigned Numbers Authority, Jon Postel, USC/ISI, 4676 Admiralty Way, Marina del Rey, DA 90292, 1984. http://info.internet.isi.edu/in-notes/rfc/files/rfc896.txt.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....or measure of load in TCP networks is the number of simultaneously active connections sharing a bottleneck, and that congestion amounts to loss rates high enough to force TCPs into timeout. An early definition of congestion in TCP networks focused on a phenomenon called congestion collapse [40]. As the load on the network increased, the throughput, queuing delay, and loss rate also increased. Increases in loss and delay caused increases in retransmission rates. These retransmissions themselves increased loss and delay, as well as consuming bandwidth to no useful purpose. The result was ....

John Nagle. Rfc896: Congestion control in ip/tcp internetworks. Technical report, Internet Assigned Numbers Authority, Jon Postel, USC/ISI, 4676 Admiralty Way, Marina del Rey, DA 90292, 1984. http://info.internet.isi.edu/in-notes/rfc/files/rfc896.txt.

....and reorder packets) The ability of TCP to adapt to networks of various characteristics and computer systems of various processing power makes TCP an important component in the fast expansion of the global Internet. The original definition of TCP appears in RFC 793 [14] Many researchers [2, 7, 8, 9, 11, 12, 18, 19] have identified problems and weakness of the protocol, and proposed solutions. RFC 1122 [1] updates and supplements the definition; to meet the TCP standard, an This workwas supported in part by a fellowship from UniForum. This paper was published in the proceedings of USENIX Summer 1994 ....

....hand, if the sender uses a smaller RTO value than the actual packet round trip time (RTT) unnecessary retransmissions occur. Moreover, if the packet round triptime increase is due to network congestion, unnecessary retransmissions make the situation even worse and may lead to congestion collapse [12]. On the other hand, if the sender uses a larger RTO value, a lost packet causes the sender to wait longer than necessary, thus degrading throughput. The calculation of the RTO value originally suggested in RFC 793 is now known to be inadequate and has been replaced. RFC 1122 specifies the new ....

John Nagle. RFC-896: Congestion Control in IP/TCP Internetworks. Request For Comments, January 1984. Network Information Center.

....data, the result is a stream of one octet data segments. To better utilize network resources, a TCP tries to buffer segments that are small compared to the size of TCP and IP headers. However, to avoid deadlock, TCP must not buffer a data segment that needs immediate delivery. Nagle s algorithm [13] provides a simple solution to the dilemma: if there is unacknowledged data (i.e. the connection is not idle) TCP buffers all data (even if the PUSH bit is set) until TCP can send an MSS segment or until all the outstanding data has been acknowledged [3, 13] Note that Nagle s algorithm also ....

....immediate delivery. Nagle s algorithm [13] provides a simple solution to the dilemma: if there is unacknowledged data (i.e. the connection is not idle) TCP buffers all data (even if the PUSH bit is set) until TCP can send an MSS segment or until all the outstanding data has been acknowledged [3, 13]. Note that Nagle s algorithm also provides sender side SWS avoidance. SunOS 4.1.1 Sender s send sequence space Data sent but not ACKed Usable window The offered receive window (Receiver s available receive buffer space) Figure 4: Illustration of a sender s usable window TCP uses the following ....

John Nagle. RFC-896: Congestion Control in IP/TCP Internetworks. Request For Comments, January 1984. Network Information Center.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC