!!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!! !!!
-------------------------------------------------------------------------------
                                       
                                       
                   B O S T O N   U N I V E R S I T Y
                                       
                      Computer Science Department
                                       
                          C O L L O Q U I U M
                                       
                 QoS-aware Network Resource Management


		 	    Ibrahim Matta


                      College of Computer Science
		        Northeastern University


 			  Wednesday, March 24th
                                       
                                11:00am
                       (Coffee served at 10:45am)

                                       
                         Seminar Room / MCS 135
                                       

-------------------------------------------------------------------------------


    Advanced networked applications, such as collaborative systems and
    interactive distributed simulations, require that the underlying
    network be able to satisfy Quality-of-Service (QoS) constraints, for
    example a delay bound on message delivery.  To effectively support
    such demanding applications, the network has to employ resource
    management protocols that are efficient, fair and QoS-sensitive.
    Furthermore, such protocols have to be developed in an integrated
    fashion, that is, the interplay among them has to be considered.
     
    In this talk, we discuss these and other requirements in the design of
    QoS-aware resource management schemes. This naturally leads to
    multi-constrained optimization problems for which exact solutions are
    prohibitively expensive to obtain. Thus, solutions are usually based
    on heuristics. We present fast and near-optimal heuristics for
    building minimum-cost delay-constrained multicast and unicast
    communication paths.  Our multicast path construction is parameterized
    by the requested QoS delay bound. Our unicast algorithm restricts the
    search space so as to speed up the path computation.

    QoS-aware resource management schemes should also result in globally
    optimal and fair allocations.  We present a new load-profiling
    distribution strategy that proactively improves fairness by increasing
    the likelihood of successfully allocating resources to future
    bandwidth-intensive demands. Finally, we discuss requirements of
    scalability and cooperative/integrated QoS management, and the need
    for fast and accurate performance prediction tools. We present such a
    tool and its embodied approximations. We used the tool successfully to
    solve dynamic flow models of reservation-oriented networks.


    Host: Azer Bestavros (best@cs.bu.edu)

-------------------------------------------------------------------------------
For colloquium info, including directions, see http://cs-www.bu.edu/colloquium
-------------------------------------------------------------------------------