Yusuf Goolamabbas

Darin ‘Necko God’ Fischer mentioned to me via email that Bug 220941 would be a good bug to fix for better performance of Mozilla over wireless links. He also mentioned that it’s one of the more difficult bugs to fix hence could be one of the reason the bug hasn’t caught much attention.

Speaking of wireless, in the current Linux 2.6.x kernel (as of 2.6.9), there are now 3 congestion control algorithms available.

• TCP Vegas
• TCP Westwood
• BIC TCP (default as of 2.6.9)

My understanding of TCP Westwood is that it is a sender-side modification of the TCP congestion window algorithm that improves upon the performance of TCP Reno in wired as well as wireless networks.

From the abstract of the ACM-Mobicom 2001 paper I gathered this.

Unlike TCP Reno, it doesn’t blindly halve the congestion window after receiving three duplicate ACK’s. Instead it uses the current effective bandwidth (TCPW continuously measures the TCP source rate) to set a slow start threshold. In wireless links, one could have sporadic losses due to radio problems and TCP Reno would interpret this as congesion which would lead to unneccasary window reduction.

So with a lot of cellphone manufacturers looking at Linux on the cell-phone, will they tune the stack for TCP Westwood. Not sure if other OS’s like Solaris 10 support different congestion control algorithms. Can one set congestion algorithms on a per-route basis ?