Kyriakos Manousakis

Applying simulated annealing for domain generation in ad hoc networks

If heterogeneous ad hoc battlefield networks are to scale to hundreds or thousands of nodes, then they must be automatically split into separate network domains. Domains allow routing, QoS and other networking protocols to operate on the fewer nodes. This division greatly reduces the overall overhead (e.g., routing overhead with n nodes that goes from O(n/sup 2/) to O(nlogn) and allows protocols to be tuned to more homogenous conditions (K. Manousakis et al., 2002). Domain generation (or clustering) can be done using either the local or global information. The two approaches are complementary since local domain generation reacts faster, requires less overhead, and is more robust while the global domain generation provides better overall domains. While the most existing work has concentrated on the local distributed solutions, this paper reports on the new global domain generation techniques. In particular we concentrate on the design of good cost functions and efficient optimization algorithms. We show that the simple intuitive cost functions do not produce good domains; rather we need complex functions with multiple parameters depending on the design goals (e.g., low overhead or low delay). Although existing optimization algorithms are too slow to be useful in a large dynamic network, we show that a modified simulated annealing algorithm with well chosen cooling schedule, state transition probabilities, and stop criteria produces good quality domains in acceptable time.

Network and domain autoconfiguration: a unified approach for large dynamic networks

Configuration management is critical to correct and efficient operation of large networks. Where the users and networks are dynamic and ad hoc, manual configuration quickly becomes too complex. The combination of the sheer number of nodes with heterogeneity and dynamics makes it almost impossible for the system administrator to ensure good configuration or even correct operation. To achieve the vision of pervasive computing, nodes must automatically discover their environment and self-configure, then automatically reconfigure to adapt to changes. Protocols such as DHCP, DDNS, and mDNS provide some degree of host autoconfiguration, but network administrators must still configure information such as address pools, routing protocols, and OSPF routing areas. Only limited progress has been made in automating the configuration of routers, servers, and network topology. We propose the first unified attempt to combine both self-configuration of much of the host, router, and server information with automatic generation and maintenance of hierarchy under the same algorithmic framework. Testbed implementations show the approach is practical, while simulations reveal its scalability, rapidity, and efficiency with respect to network performance.

Using multi-objective domain optimization for routing in hierarchical networks

Network hierarchy makes network protocols more scalable and robust, but also makes the network more complex and reduces performance. With routing protocols, hierarchy reduces overhead, routing table size, and convergence time, but can also cause sub-optimality (stretch) of the routing path length compared to the flat networks. With OSPF, for example, adding routing areas, with route aggregation done in area border routers, can significantly reduce link state advertisements, link state database and convergence time, but can also increase interarea path length. Existing research has produced many excellent intra- and inter-domain routing schemes and different proposals for aggregation at border routers. As important, however, is the design of the routing hierarchy itself. This paper presents quantified comparisons of different hierarchical construction techniques for a simple hierarchical routing protocol in a 100 node network. When the hierarchy does not take into account routing path length suboptimality, we show the potential for significant stretch (e.g., on average more than doubling the shortest path for nodes under 6 hops apart). When we use multi-objective optimization, that includes the goal of minimizing stretch, the stretch is significantly reduced (e.g., reducing the stretch by approximately 50% for the above example). In large or dynamic environments, a modified simulated annealing is able to produce optimized hierarchies quickly by trading a small loss in optimality for a large reduction in optimization time. The paper analyzes the choice of multi-objective cost function in this environment and concludes that simpler functions produce the best overall results.

Clustering for transmission range control and connectivity assurance for self configured ad hoc networks

Ad hoc networks have been the new networking technology trend because of their promising characteristics. These characteristics fit better the requirements of todays army and the needs of todays commercial world. Most of the applications that have been referred in the bibliography assume devices that are of finite power. The latter is true because of the existing technology (web enabled cell phones, PDAs, laptops, PPCs). The goal for ad hoc networks is to accommodate light weight, battery powered portable devices. Because of the finite power limitation, we have to design efficient ways to use the existing power. A first step towards efficient utilization of power is to eliminate its unnecessary usage wherever possible. For a networking device, significant part of power is consumed for communications and more specifically for transmissions. In this work we focus on eliminating the transmission power consumed into the network subject to the networks connectivity. The assurance of network connectivity is essential in this problem since the usage of limited transmission power from the nodes may result in a partitioned network. We provide solutions to the efficient usage of transmission power by clustering the nodes based on their proximity and ensuring the intracluster and intercluster connectivity of the network. The algorithms presented in this paper achieve their objective as the collected results from their simulation show.

Multi-metric Energy Efficient Routing in Mobile Ad-Hoc Networks

Increasing network lifetime by reducing energy consumption across the network is one of the major concerns while designing routing protocols for Mobile Ad-Hoc Networks. In this paper, we investigate the main reasons that lead to energy depletion and we introduce appropriate routing metrics in the routing decision scheme to mitigate their effect and increase the network lifetime. For our routing scheme, we take into consideration multiple layer parameters, such as MAC queue utilization, node degree and residual energy. We integrate our multi-metric routing scheme into OLSR, a standard MANET proactive routing protocol. We evaluate via simulations in NS3 the protocol modifications under a range of different static and mobile scenarios. The main observations are that in static and low mobility scenarios our modified routing protocol leads to a significant increase (5%-20%) in network lifetime compared to standard OLSR and slightly better performance in terms of Packet Delivery Ratio (PDR).

A Robust, Distributed TGDH-based Scheme for Secure Group Communications in MANET

Securing multicast communications in Mobile Ad Hoc Networks (MANET) is considered among the most challenging research directions in the areas of wireless networking and security. MANET are emerging as the desired environment for an increasing number of commercial and military applications, addressing also a growing number of users. Security on the other hand, is now an indispensable requirement for these applications. However, the limitations of the dynamic, infrastructure-less nature of MANET impose major difficulties in establishing a secure framework suitable for such services. The design of efficient key management (KM) schemes for MANET is of paramount importance, since the performance of the KM functions imposes an upper limit on the efficiency and scalability of the whole secure group communication system. In this work, we contribute towards this direction by extending TGDH to a novel distributed scheme: DS-TGDH. Our aim is to modify TGDH to: a) be feasible in the most general resource-constrained MANET where no nodes with special capabilities exist, b) produce considerably lower overhead for the network nodes involved, c) handle disruptions with low cost. We consider the underlying routing protocol in our design, and we apply a distributed TGDH version over a robust schedule, optimizing parameters of interest. We focus on the design and analysis of the stealthy TGDH and compare it with the original.

Design framework for hierarchy maintenance algorithms in mobile ad hoc networks

Domain autoconfiguration techniques allow the quick formation of highly optimized hierarchies that greatly enhance network scalability and overall performance. For example, instead of producing a simple two level hierarchy based only on topology, the optimization can produce multi-level hierarchies that take into account factors such as mission goals and predicted node/link heterogeneity. However, in dynamic networks, such as expected in the future military networks, these highly optimized solutions degrade very quickly. Indeed, if we use standard local maintenance algorithms that do not align well with the optimization goals, then the performance can reach the level of a suboptimal solution in less than two minutes. This paper proposes a taxonomy of local maintenance algorithms into four basic classes and quantifies the performance benefits of using representative approaches that act in accordance with the optimization goals.

Routing through an integrated communication and social network

This paper explores the robust routing of messages among individuals. Traditional routing assumes individuals provide messages to a device connected to a communications network that assumes all responsibility for message delivery. Although each individual may have links to multiple communication devices (office computer, PDA, cell phone), messages are delivered only if there is an end-to-end communication path between communication devices available to each individual. To improve robustness of communication, especially in dynamic ad hoc military networks, this paper models a novel routing paradigm using an integrated communication and social network. The understanding is that individuals can and do route messages through a social network in conjunction with the communication network. An example of this is an individual asking another in his immediate social network to place a call on his behalf when the official communication system is not convenient or is unavailable. We show that it is possible to route messages through the integrated social and communication network by: a) using the ORA social analysis tool to select normalized costs for the social and communication network links (e.g., to reflect the link delay, quality or robustness), and b) using the MONOPATI communication design tool to model the integrated socio-communication network as a graph and performing QoS routing. Results show that the robustness of message delivery can be improved by 5X through this joint routing, without unnecessary impacts on end to end latency.

Rate of degradation of centralized optimization solutions and its application to high performance domain formation in ad hoc networks

Future military systems such a FCS and WIN-T require a robust and flexible network that supports thousands of ad hoc nodes; therefore, we must ensure the scalability of networking protocols (e.g., routing, security and QoS). The use of hierarchy is a powerful solution to the scaling problem, since it allows networking protocols to operate on a limited number of nodes, as opposed to the entire network. We have proposed an automated solution to dynamically create and maintain such hierarchy based on a combination of global optimization algorithms (K. Manousakis et al., 2004) and local distributed maintenance protocols (R. Morera et al., October 2003). Global optimization clearly improves performance in a static network but it is unclear how effective it is in a dynamic ad hoc environment. In this paper, we analyze how the hierarchy created deteriorates from the optimal as network conditions change. We show that the fragility of the optimization depends on the particular cost function and the number of metrics that change. More important, we show, for the first time, that global optimization can remain effective for long periods with good cost functions, even in large dynamic ad hoc networks (where metrics may change rapidly due to node mobility and links making and breaking). This result shows that, with fast optimization algorithms such as modified simulated annealing (K. Manousakis et al., 2004), future military systems can use global optimization to autoconfigure domains to significantly improve performance. We also show that local maintenance protocols support the global optimization mechanisms by extending the time the hierarchy remains feasible.

Scalable and robust reliable multicast for satellite networks

The evolution of satellite applications has created new traffic patterns on satellite links. That is because the number of users demanding service from satellite links has increased tremendously. The result is a more sophisticated use of resources. One way of accomplishing that is by multicasting, which is a natural extension of the broadcast nature of the satellite medium. Numerous existing satellite applications demand reliable delivery of the transmitted data, which is opposed to the unreliable nature of multicast transmission due to the error prone satellite medium. We have designed a number of reliable multicasting protocols which are based on forward error correction (FEC) and air caching. The combination of those two techniques results in a significant boost in the performance of reliable multicasting protocols.