Abstract
Open shortest path first (OSPF) protocol is commonly used as an interior gateway protocol (IGP) in MPLS and
generalized MPLS (GMPLS) networks to determine the topology over which label-switched paths (LSPs) can be
established. Traffic-engineering extensions (network states such as link bandwidth information, available wavelengths,
signal quality, etc) have been recently enabled in OSPF (henceforth, called OSPF-TE) to support shortest path first (SPF)
tree calculation upon different purposes, thus possibly achieving optimal path computation and helping improve resource
utilization efficiency. Adding these features into routing phase can exploit the OSPF robustness, and no additional
network component is required to manage the traffic-engineering information. However, this traffic-engineering
enhancement also complicates OSPF behavior. Since network states change frequently upon the dynamic trafficengineered
LSP setup and release, the network is easily driven from a stable state to unstable operating regimes. In this
paper, we focus on studying the OSPF-TE stability in terms of convergence time. Convergence time is referred to the
time spent by the network to go back to steady states upon any network state change. An external observation method
(based on black-box method) is employed to estimate the convergence time. Several experimental test-beds are
developed to emulate dynamic LSP setup/release, re-routing upon single-link failure. The experimental results show that
with OSPF-TE the network requires more time to converge compared to the conventional OSPF protocol without TE
extension. Especially, in case of wavelength-routed optical network (WRON), introducing per wavelength availability
and wavelength continuity constraint to OSPF-TE suffers severe convergence time and a large number of advertised link
state advertisements (LSAs). Our study implies that long convergence time and large number of LSAs flooded in the
network might cause scalability problems in OSPF-TE and impose limitations on OSPF-TE applications. New solutions
to mitigate the s convergence time and to reduce the amount of state information are desired in the future.
