This paper proposes a completely redesigned protocol that helps to manage congestion. Instead of using packet drop as a binary indicator, the paper extends on previous Explicit Congestion Notification (ECN), and use explicit headers from end hosts and precise feedback from the gateway to achieve congestion control. This is also designed to solve the instability and oscillation problem found in some previous congestion avoidance schemes based on TCP.
Key ideas:
1. decouple the fairness control from utilization control.
2. Error loss != congestion loss , especially true for wireless links and links with long delays
Cons:
1. Deployment will need to be a complete overhaul.
2. Leaving too much power in the hand of the end hosts. The control state contained in the packet can be modified to benefit the end hosts in several ways.
Subscribe to:
Post Comments (Atom)
1 comment:
So doesn't really require a complete redesign, as XCP can co-exist with TCP. Also everything else we have read thus far pretty much depends on sources being loss-responsive in order for congestion control to work. What else could you possibly do?
Post a Comment