Today all Internet routers are built with buffers that hold packets in
times of congestion. These buffers can typically store between 250ms to one
second worth of data. According to a widely used Òrule-of-thumbÓ, a link
needs a buffer of size B = RTT x C, where RTT is the average round-trip
time of a flow passing across the link, and C is the data rate of the
link. For example, a 10Gb/s router linecard needs approximately 250ms x
10Gb/s = 2.5Gbits of buffers; and the amount of buffering grows linearly
with the line-rate. Such large buffers are challenging for router
manufacturers, who must use large, slow, off-chip DRAMs. They also lead to
high latencies in case of congestion. Recent research suggest that the
rule-of-thumb (B = RTT  x C) is now outdated and incorrect for routers
serving highly aggregated traffic. According to these new results, a link
with n flows requires no more than B = (RTT x C)/ sqrt(n), for long-lived or
short-lived TCP flows. These lower buffer requirements have been verified
in simulation and laboratory experiments as well as on real networks with
live traffic. The consequences on router design are enormous: a 2.5Gb/s
link carrying 10,000 flows could reduce its buffers by 99% with negligible
difference in throughput, and a 10Gb/s link carrying 50,000 flows requires
only 10Mbits of buffering, which can easily be implemented using fast,
on-chip SRAM.