Preparation for Data Center TCP (DCTCP) recitation
DCTCP customizes the TCP congestion control algorithm for data centers. It
leverages the Explicit Congestion Notification (ECN) to obtain an early
congestion feedback from routers/switches, before the queue drops packets.
Further, DCTCP provides a smooth reaction to congestion, i.e., when congestion
is limited, it reduces its congestion window by a small amount. In contrast,
when congestion is severe, it reduces its congestion window by a large amount.
Read the first 3 sections of the paper and try to understand how DCTCP compares
to TCP. Skim the Results section (i.e., section 4) to check that the empirical
results match your expectations. In particular check Figures 15 and 19, which
show the queue occupancy as a function of time, and number of sources.
The following slides summarize the paper and could be helpful in
understanding the content:
Consider the following questions:
- How do the requirements of congestion control change as
we move from the Internet to Data Centers?
- What is the incast problem?
- What particular features in the design of DCTCP cause the queues to be
smaller than the case where the network uses TCP?
- What happens if we have two flows that share the same bottleneck but
one of them uses TCP and the other uses DCTCP? In particular, which flow
do you expect to have a higher throughput and why?