 
Summary: Review of "Congestion Control for High BandwidthDelay
Product Networks" by Katabi et al.
Sreenivasa Pavan Kuppili
February 23, 2007
TCP becomes inefficienct and oscialltory (irrespective of the queueing scheme) in the presence of high
bandwidthdelay links. So this paper comes up with a protocol which is efficient, fair, scalable, and stable as
the bandwidthdelay product increases. The routers give explicit and precise feedback to the senders, telling
the sender how to change the cwnd value (feedback of the bottleneck link prevails). The paper puts forth the
idea of decoupling utilization control from fairness control. Utilization control tries to optimize the aggregate
throughput. To calculate the desired amount of throughput change in the next d units of time, it uses the
equation = .d.S  .Q, where and are constant parameters, d is the mean RTT of the flows through
the link, S is the spare bandwidth, and Q is the persistent queue size. So the efficiency controller using
an MIMD approach and tries to quickly attain the spare bandwidthdelay product. This MIMD approach
increases the efficiency in links with a highbandwidth delay product. The efficiency controller also tries to
drain the persistent queues. This proactive reduction of queue size results in small queues and very small
packet losses. The fairness controller distributes this aggregate throughput among the various flows in an
AIMD way to attain fairness. If > 0, it is allocated so that the throughputs of all the flows increase by the
same amount in the same time (flows with larger RTT increase their cwnd quickly, and TCP's bias against
flows with large RTT's is avoided). If < 0, it is allocated so that the decrease in throughput is proportional to
the throughput. The fairness controller can naturally be extended to perform service differentiation. Further,
