Aron Szentesi

On bandwidth efficiency of the hose resource management model in virtual private networks

The hose resource provisioning model promises to provide an easy-to-use characterization framework for virtual private network service offerings. Significant research effort has recently been spent on proposing new algorithms for provisioning cost-optimal networks specified according to this new model. However, a detailed comparison of the bandwidth requirement for networks designed based on the hose model and networks designed based on the traditional pipe model has not been performed. The first contribution of this paper is a detailed comparison of the bandwidth needs of the two models assuming a range of network sizes and network topologies. This numerical evaluation required efficient calculation methods for determining resource allocation based on the hose model parameters, therefore, a linear programming based formulation is also presented for this purpose. The second contribution is the calculation of a lower bound for the hose based realization. This lower bound is very useful in evaluating the two models given that the problem of provisioning a minimal cost network based on the hose model specification can only approximately be solved in polynomial time.

Inverse optimization in high-speed networks

A general approach is presented for handling the following inverse optimization problem: given solutions to each member of a family of combinatorial optimization tasks on a common underlying set, find a positive linear objective function (weighting) on the common underlying set that simultaneously makes each solution optimal in its own optimization task. Our motivation stems from the inverse shortest path problem that is made practically important in high-speed telecommunication networks by the Asynchronous Transfer Mode Forums Private Network-Network Interface architecture, in which route finding can be based on administrative weights. Different variants of the problem are investigated, including uniqueness requirements and reserve routes.

Minimizing re-routing in MPLS networks with preemption-aware constraint-based routing

In this paper, we study the effect of distributed constrained shortest path first (CSPF)-based path selection on the dynamics of label switched path (LSP) preemption. We propose new CSPF algorithms for minimizing preemption of lower priority LSPs without requiring any enhancements to the recently proposed link-state parameters. The difference between priority-based path selection methods and previously proposed CSPF methods lies in the way the selection is done among equal cost shortest paths. Our priority-aware CSPF algorithms decrease the number of preempted lower priority LSPs while the LSP setup success ratio is basically the same for all methods.

Planning of tree-topology UMTS terrestrial access networks

In this paper we propose a heuristic planning algorithm for optimizing tree-topology access networks. Our suggested method is able to plan a multi-constrained and capacitated subnetwork tree with a previously dedicated root node. The algorithm is a combination of an adaptive version of the simulated annealing meta-heuristic and a local improvement strategy. We use the proposed method to solve the optimization problem of UMTS (universal mobile telecommunications system) access networks. Beside the description of the model and the algorithm we present a detailed performance analysis of our proposed method using different network configurations and traffic demands. Tests show that our proposed algorithm provides solutions very close to the global optimum and because of its computational efficiency it is usable for planning practical size networks. It is also important to note that our method can be easily adapted for similar planning tasks in case of many other technological backgrounds.

On-demand optimization of label switched paths in MPLS networks

Multiprotocol label switching (MPLS) technology provides efficient means for controlling the traffic of IP backbone networks and thus, offers performance superior to that of traditional interior gateway protocol (IGP)-based packet forwarding. However, computation of explicit paths yielding optimal network performance is a difficult task; several different strategies can be thought of. We propose a novel approach for the optimal routing of new label switched paths (LSP). Our method is based on the idea that we allow the re-routing of an already established LSP when there is no other way to route the new one. Our optimization algorithm is based on an integer linear programming (ILP) formulation of the problem. We show by numerical evaluation that our method significantly improves the success probability of setting up new LSP by extending the widely known constrained shortest path first (CSPF) algorithm with the ILP-based re-routing function. Although the task itself is NP-hard, our heuristic method provides efficient solution in practical cases. The proposed algorithm is aimed to be implemented primarily in a network operation center. However, with the restriction that only LSP starting from the optimizing label edge routers can be re-routed, the algorithm can be implemented directly in edge routers themselves in a distributed fashion.

Analysis of link cost functions for PNNI routing

Abstract The emerging environment of ATM networks with unpredictable traffic patterns and ever-changing network configurations create the need for using dynamic routing methods. Dynamic routing basically means that path selection continuously adapts to the state of the network. The routing framework of the Private Network-to-Network Interface (PNNI) specification achieves this by using link-state information gathered from the network. One possible approach for selecting paths is to use a shortest path algorithm (e.g., Dijkstras) with link costs generated from the link load information. This mapping is realised by the so-called link cost function. We investigated a number of such link cost functions with simulation. We developed two novel link cost functions and compared them with those found in the literature. The primary basis of comparison was the total throughput of the networks. Besides this, we also investigated such issues as imposed signalling load and robustness.