ppt课件-chapter5end-to-endprotocols-ualr.pptVIP

In a the sample RTT is longer than the actual and in b it is smaller than the actual. These solutions were added in 1987. This institutes an exponential back off each time a retransmission occurs. This is similar to the exponential backoff in Ethernet. In Ethernet the back off is used to resolve collisions. In TCP the assumption is that most retransmissions are due to congestion. Aggressive retransmission could exasperate a congestion problem and this procedure was designed to help alleviate that issue. At this point the traffic on the internet was rapidly increasing and congestion ‘storms’ were becoming more common . A strategy to allow TCP to continue to function with varying congestion levels was a high priority. The EstimatedRTT, used to set the actual timeout, must be designed to both keep the connection going during congestion and also limit retransmissions to the minimum necessary to remain connected. This is to avoid the data in the retransmissions from adding to the existing congestion, making it worse or even causing collapse Following Karn/Partridge, Jacobson and Karels proposed using a statistical calculation to set the timeout. μ is set to the running average of RTT and the deviation is set to show the current difference from the mean. As the deviation grows its effect on the timeout increases, making it more sensitive to congestion related timeouts. With low deviation the timeout will hover around the EstRTT calculated using the send and ACK times. By adding extensions using header options they can be negotiated during the session setup and used if they are present at both ends without requiring any customization. The first extension is the adding of a 32 bit system clock synchronized timestamp to the header. This enables very accurate RTT calculations, making the procedures described earlier to reduce the affect of congestion, more effective. The second extension is to make the effective sequence number 64 bits long by using the 32 bi