All problems equally weighted (1/3 each)


Problem #1:  TCP on top of ATM

Consider an ATM network which supports TCP traffic
using AAL5.  For each TCP connection, a VC is set up
in the ATM net.  However, no bandwidth is allocated,
since the user has no way to estimate/predict the
usage parameters (average, peak, burstiness).  For
congestion control, in view of the lack of input
rate control, hop-by-hop backpressure using
a credit scheme is applied
along the VC (like in an X.25 network).

Questions:

(a)     assume that a TCP connection can inject a
        peak rate of 10MBBPS into the net (of
        course, average rate will be much lower).
        Consider an ATM trunk 1000 miles long.  How
        large a credit (in ATM cell units) must be
        implemented in order to support data
        transfer at peak TCP rate?

(b)     How large a buffer must be provided for each
        VC in order to avoid buffer overflow?

(c)     Discuss the implementation problems (buffer
        allocation, control overhead, etc.) posed
        by the VC hop-by-hop flow control.  For
        example, how would you go about implementing
        this scheme in the Knockout switch?  How
        would you integrate on the same ATM trunk
        TCP traffic and real time traffic (with
        bandwidth allocation and input rate
        control)?
        
        
Problem 2: Multimedia with adjustable rate

Consider a multimedia application environment
supported by an ATM network.  Assume that each
multimedia connection has an adjustable rate in the
range 1-10MBPS (where 1MBPS corresponds to lowest
quality, 10 MBPS corresponds to best quality).  We
want to provide the best possible quality of service
to users; yet, we must allow users to share
resources in a fair way.  For example, if there are
only 3 connections sharing a 150 MBPS trunk, they
should get the full 10 MBPS rate each.  However, as
the load on the trunk increases to, say, 100
connections, the rate on each connection should drop
to 1.5 MBPS.  Clearly, the conventional static
bandwidth allocation at call set up time will not
work, since it will lead to unfairness.  So, a new,
dynamic bandwidth allocation scheme must be
designed.  To simplify matters, we assume that calls
are routed on shortest paths.  Furthermore, each
source is periodically informed of the residual
bandwidth still available on the path to each
destination (note that the residual bandwidth
changes dynamically as calls are established and
cleared).

(a)     discuss schemes for effectively propagating
        residual bandwidth info to sources (eg, in
        the context of a Bellman-Ford type routing
        alogorithm)

(b)     propose a distributed, dynamic rate
        adjustment algorithm, in which each source
        adjusts its rate according to the residual
        bandwidth value advertised by the network.
        You must show that your method is fair (ie,
        all users sharing a bottleneck transmit at
        the same rate).  
        
Problem 3: optical networks     
        
Consider the design of an all optical network using
multifiber cables.  It has been suggested that
multiple fibers in a cable play the same role as
multiple wavelengths in a single fiber.  That is, we
can set up and end-to-end connection by selecting
one of the fibers in the cable, or by selecting one
wavelength in the single fiber situation.  Elaborate
on this analogy, and discuss the pros and cons of
using "fiber switching" vs. "wavelength switching".