Sung-eok Jeon

The optimal connection preemption algorithm in a multi-class network

In an integrated network in which multiple classes with different priorities exist, when the network does not have enough unused bandwidth, connection preemption algorithms (CPA) play an important role in accepting a new session with a high priority by preempting lower priority flows already admitted. Although there are lots of studies about which connections to preempt to optimize the preemption factors (the priority of connections, the bandwidth to be preempted, and the number of connections to be preempted), existing CPAs are only suboptimal from the point of view of the preemption factors because of the computational complexity. The connection preemption problem in a centralized fashion is proved to be NP-complete. To avoid the complexity of the centralized scheme, decentralized/distributed algorithms are proposed. However their solutions are optimal only at the hop level. In order to avoid the priority order problem and to minimize the number of preempted and rerouted sessions, we propose to order the priority of the preemption factors in a new way. We also propose an optimal connection preemption algorithm which optimizes the newly ordered preemption factors in the connection preemption events. The proposed algorithm is the first optimal algorithm that provides a guideline about the upper bound of the computational complexity of the optimal CPAs.

Nearly optimal distributed configuration management using probabilistic graphical models

This work studies distributed configuration management of large wireless sensor networks, where management objectives are achieved by local cooperation of individual nodes. Specifically, we study when distributed configuration management is nearly optimal, and how to obtain a nearly-optimal configuration through decentralized adaptation. We first derive a spatial network model that is determined by internal network characteristics and management requirements. We next show that a sufficient condition for distributed configuration management to be nearly-optimal is that the spatial network model belongs to a class of coupled Markov random fields also known as random-bond ising model. Such graphs possess a cross-layer spatial Markov property. We specify the sufficient conditions for the nearly-optimality under different channels and density of nodes. We derive a nearly-optimal distributed algorithm using the probabilistic inference based on the derived network model. The algorithm is applied to spatial-reuse TDMA which configures a logical topology

Formulation and optimization of the connection preemption problem

In the literature, the connection preemption policy is proposed for the resource allocation method in the prioritized multi-class networks. However, the connection preemption problem that is defined in the literature does not consider the rerouting of the preempted connections. When some connections are preempted at a connection preemption event, a part of the preempted connections will be rerouted with their own preemption priorities. Therefore, each rerouted connection setup can also trigger new connection preemption events. In this work, we propose to extend the existing connection preemption problem to include the subsequent connection preemption events (triggered by new or rerouted connections). We call the new extended problem the extended connection preemption problem. We formulate the existing connection preemption problem and the extended connection preemption problem with their decision variables and cost function. This is the first work that formulates the connection preemption problem in terms of its cost function. And we also analyze how to optimize the cost function of the extended connection preemption problem.

Path selection with class distribution information in the integrated network

In an integrated network, QoS routing plays an essential role in supporting various classes with different priority and quality-of-service (QoS) characteristics. The path selection procedure is an important part of the QoS routing scheme. The existing path selection algorithms do not consider the class distribution information of residual bandwidth (RB-CDI). As a result, these algorithms may cause the sessions of higher priority classes to be preempted instead of those of lower priority classes. We propose a new and straightforward path selection algorithm to solve this problem. The proposed algorithm uses not only the information of residual bandwidth amount but also RB-CDI. We evaluate the proposed algorithm by comparing with existing path selection algorithms. The performance evaluations are done in various situations. We use the throughput of each class and the link utilization of the network as evaluation metrics.

Topology aggregation: Merged-star method for multiple non-isomorphic topology subgraphs

In a hierarchical network, each subnetwork advertises its topology information (i.e., topology graph) to the other subnetworks. Lots of work has been done to minimize the amount of advertised topology information. With the existing methods, however, the amount of advertised information linearly increases to the order of the number of link-state parameters. To solve this problem, using graph theory, we show how to convert multiple non-isomorphic topology-advertising subgraphs into an isomorphic graph.

Role of machine learning in configuration management of ad hoc wireless networks

In this work, we show that machine learning, e.g., graphical models, plays an important role for the self-configuration of ad hoc wireless network. The role of such a learning approach includes a simple representation of complex dependencies in the network and a distributed algorithm which can adaptively find a nearly optimal configuration.

Topology aggregation method for multiple link parameters

In a hierarchical network, each subnetwork advertises its topology information to the other subnetworks. Many studies have been done to minimize the amount of advertised topology information. The existing works assume a logical link can be represented with only a few parameters. However, for the QoS and policy-based networks, more parameters are used to represent the state of a link (e.g., cost, policies). For these networks, with the existing methods, the amount of advertised information linearly increases to the order of the number of link-state parameters. To solve this problem, this work proposes a new topology aggregation method.

Concurrent Failures and Redundant Protection Problem in Hierarchical GMPLS Networks

The generalized multiprotocol label switching (GMPLS) networks attain a hierarchical structure, andeach layer maintains an independent protection mechanism, resulting in redundant protection. The common pool method provides a basic approach to solve the redundant protection problem. Although the common pool method is simple and robust, however it can fail in some cases. This work first shows that none of the prior work satisfies both redundant protection and two links failure problem simultaneously. Moreover, this work also presents a new type of two links failure problem (i.e., the failure of two links at two different layers), which can happen frequently and the common poolmethod cannot cope with. To solve the proposed two links failure problem, while minimizing the cost of redundant protection problem, this work proposes a new protection scheme for hierarchical GMPLS networks.

Redundant Protection Problem in the Hierarchical GMPLS Networks

The generalized multiprotocol label switching (GMPLS) networks possess a hierarchical structure, and each layer has its own independent protection mechanism. Thus, there is redundant protection due to the survivability mechanism at each layer. Among many proposed solutions on the redundant protection problem, the common pool method is considered as the most representative one. However, the common pool method cannot cope with two link failures. With a proposed method, the common pool method can be extended to cope with two link failures with a small cost

Near-Optimal algorithm for self-configuration of ad-hoc wireless networks

To quantify the goodness of a configuration, we develop a probabilistic model of network configuration. A probabilistic graph then represents the statistical dependence in network configuration, and shows that self-configuration can be optimized if the graph has nested local dependence.