Title:
Topological Design of Multiple Virtual Private Networks Utilizing Sink-Tree
Paths
When: Wednesday, July 28, 2004, 10:30-11:30
AM
Where: Room 503 IS Building
Who: Anotai Srikitja
Committee: Dr. David Tipper, University
of Pittsburgh , Committee Chair; Dr. Richard Thompson,
University of Pittsburgh Committee Member ; Dr. Prashant
Krishnamurthy, University of Pittsburgh Committee Member;
Dr. Bryan Norman, University of Pittsburgh Committee
Member; Dr. Peter Steenkiste, Carnegie Mellon University
Committee Member
Abstract: With the deployment of MPLS
over a core IP backbone, it is possible for a service
provider to built Virtual Private Networks (VPNs) supporting
various classes of services with QoS guarantees. Efficiently mapping the
logical layout of multiple VPNs over a service provider network is a
challenging traffic engineering procedure. The use of sink-tree
(multipoint-to-point) routing paths in a MPLS network makes the VPN design
problem different from traditional design approaches where a full-mesh of
point-to-point paths is often the choice. The clear benefits of using
sink-tree paths are the reduction in the number of label switch paths and
bandwidth saving due to larger granularities of bandwidth aggregation
within the network.
In this thesis, the design of multiple VPNs over MPLS-like
infrastructure network, using sink-tree routing, is formulated
as a mixed integer programming problem to simultaneously
find VPNs logical topologies and their dimensions to
carry multi-service, multi-hour VPNs traffic from various
customers. Such a problem formulation yields a NP-hard
complexity. A heuristic path selection algorithm is proposed
here to scale the VPN design problem by choosing a small-but-good
candidate set of feasible sink-tree paths over which
the optimal routes and capacity assignments are determined.
The proposed heuristic has clearly shown to speed up
an optimization process and the optimal solution can
be obtained within a reasonable time for a realistic-size
network.
Nevertheless, when a large number of VPNs are being
layout simultaneously, a standard optimization approach
has a limited applicability in practice. The Minimum-Capacity
Sink-Tree Assignment (MCSTA) algorithm is, therefore,
proposed to approximate optimal bandwidth and sink-tree
route assignment for multiple VPNs within a polynomial
computational time. The MCSTA algorithm is demonstrated
to give a good approximation within a small error bound
and sometimes yields the exact solution. Lastly, the
proposed VPN design models and solution algorithms are
extended for multipoint traffic demand including multipoint-to-point
and broadcasting connections. |