Hello, During experiments with delays and packet losses I have noticed a strange behavior of the 2.2.17..19 kernels. What I observe is that one or two segments are lost due to a full buffer in a gateway (fbsd dummynet), and so subsequent segments are selectively acknowledged. After a while, the lost segments are retransmitted and properly acknowledged. However, the segments previously SACK'ed are not included in this acknowledgement, forcing retransmission of them. AFAIK, since the reciever has sent an SACK for a segment, it is received OK and probably buffered to be taken advantage of later, but in my experiments they seem to have been discarded. It would be more efficient (in terms of network utilization) to preserve them and ACK them all when the missing segment(s) arrive. For a visual description, see the zoomed tcptrace/xplot output at http://www.cse.kau.se/~di5alfs/tt/closeup.ps The xplot + pcap file is available in the 'tt' directory as well. The capture is taken on the receiver. Also there is a delay + filled router buffer between the sender and receiver, which is why there is no direct fast retransmit after the dupacks, as I understand it. Has anyone seen this before, or know what is going on? Regards, Stefan - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
