Monday, September 28, 2009

18. A Comparison of Mechanisms for Improving TCP Performance over Wireless Links

In TCP protocol, packet drops are mostly considered to be network congestion. But in wireless links, they could result from iffy link conditions. TCP doesn't tell the difference between these, and drops transmission window size before retransmitting packets, and the link speed will be going from bad to worse.

Two approaches:
  • hide non-congestion losses from TCP sender, making lossy link appears as high quality link with a reduced bandwidth. AIRMAIL, indirect-TCP, snoop protocols
  • make sender aware of wireless hops and packet losses are not because of congestion, good bandwidth
Several schemes were proposed and are compared in this paper, including
  • end-to-end retransmissions: selective ACKs (SACKs), Explicit Loss Notifications (ELN)
  • split-TCP connections: hide wireless link from sender
  • link-layer proposals: hide link losses from TCP sender by local retransmissions and forward error correction on wireless links
Benchmark was measured by throughput and goodput for wired and wireless link.

Comment: Would it be possible to apply XCP on wireless link because it has an extra field of congestion level, so that packet drops of lossy link can be distinguished from congestion?
Posted by at
Labels: ,

1 comment:

  1. Sameer AgarwalOctober 1, 2009 at 2:34 AM

    Nice idea. But the question would be how important is congestion control in wireless network. Having an XCP congestion header is considerable overhead if it is just being used to notify about loss/congestion (in which case just one bit ELN notification will do a fine job). However, if tackling congestion in wireless network is important, then a scheme having XCP like congestion header will definitely serve dual purpose.

    ReplyDelete

Subscribe to: Post Comments (Atom)