Hi,
Our case may be a bit different. Sara/NIKHEF send data out to us on a 10G line. Both Geant ports to Israel are 2.5G. Our Institute port is 1G. So the "impedance" mismatch is high and we suspect packets are dropped. The congestion algorithms only kick in when they sense dropped packets. We have fixed other transfer problems so I can't be sure the improvements are 100% due to changing the congestion algorithm. However, Onno Zweers at SARA reports an improvement (90% reduction in "FTP proxy did not shut down" errors) on the line to the Irish Tier2 after making this change. There are so many things changing beyond our control it is very hard to know what is affecting what. In any case, the bic and cubic algorithms have known bugs up to kernel version 2.6.33.
Cheers,
Lorne
|