Congestion Control for High Bandwidth-Delay Product Networks

   
  Congestion Control for High Bandwidth-Delay Product Networks
   
  This paper presents eXplicit Control Protocol (XCP). TCP is inefficient and prone to instability as the per-flow product of bandwidth-delay increases and XCP is proposed to address this problem. XCP generalizes Explicit Congestion Notification (ECN) and decouples utilization control from fairness control.
   
  The rationale behind XCP is that packet loss is a poor and imprecise signal to congestion. As congestion is not a binary value, the routers use a congestion header to inform the endpoint from the degree of congestion. Senders maintain their congestion window cwnd and round trip time rtt communicate these to routers via congestion header in every packet. The job of an XCP router is to compute feedback to cause the system to converge to optimal efficiency and min-max fairness. To compute this feedback, an XCP router uses an efficiency controller and a fairness controller. Efficiency controller (EC) uses Multiplicative-Increase Multiplicative-decrease (MIMD) to adapt the aggregate traffic rapidly to the spare bandwidth of the router. On the other hand, fairness controller (FC) uses Additive Increase-Multiplicative-Decrease (AIMD) to adjust individual feedback to achieve fairness. A fluid model of the traffic proves that system is stable.
   
  Based on several simulations, paper has compared XCP performance with TCP Reno over following queuing disciplines: Random Early Discard (RED), Random Early Marking (REM), Adaptive Virtual Queue (AVQ) and Core Stateless Fair Queuing (CSFQ). Simulation results show that as the capacity increases, TCP bottleneck utilization decreases significantly, regardless of queuing scheme. In contrast, XCP’s utilization is always near optimal of the link capacity. Furthermore, XCO never drops any packet, whereas TCP drops thousands of packets despite its use of ECN. Other comparisons have been made in other issues like performance of the system as propagation delay and number of flows increase as well as impact of short web-like traffic and fairness. All these comparisons show that XCP’s performance is significantly better than TCP. From the dynamics point of view, XCP dampens oscillations and converges to high utilization smoothly. In addition, it shows robustness to sudden increase or
 decrease in traffic demands.
   
  XCP facilitates the job of policing agents to identify and isolate misbehaving sources by its explicit feedback. The routers can send a suspect flow a test feedback to decrease its congestion window to a particular value and if the flow does not react, they can consider it as unresponsive. XCP also can be deployed gradually to substitute TCP in the present situation.
   
  In conclusion, XCP is viable and practical as congestion control scheme. It operates the network with almost no drops, and substantially increases the efficiency in high bandwidth-delay product environments.
   

                                 
---------------------------------
Want to be your own boss? Learn how on Yahoo! Small Business.
Received on Mon Oct 02 2006 - 21:32:58 EDT