Hi all, I'm running a TCP transfer with /proc/sys/net/ipv4/tcp_sack = 1 in Linux 2.4.18, and a block of segments are lost in the middle during a cold-handoff (changing IP addresses), and I'm seeing a strange behavior after fast retransmit/recovery. Specifically, TCP enters slow start rather than congestion avoidance after fast retransmit/recovery, and the window size (cwnd) is set to 2 segments rather than half of the cwnd before entering fast retransmit. I've also done a "netstat -s -p tcp" before and after the TCP transfer, and TCPTimeouts stays the same. and also looking at the sequence # vs. time plot at http://www.cs.berkeley.edu/~shelleyz/temp/roam_mipv6_100ms_sack.eps (ignore the cyan and yellow points), there was no timeout to cause TCP to go into slow start. so i'm wondering how cwnd is set when exiting fast retransmit/recovery? Any help is greatly appreciated. Thanks, Shelley - : 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
