In-Stream Data Processing for Tactical Environments

In-Stream Data Processing for Tactical Environments

Marco Carvalho (Institute for Human and Machine Cognition, USA)
DOI: 10.4018/978-1-60566-130-8.ch017
OnDemand PDF Download:


Data dissemination and information management technologies for tactical environments are quickly becoming major areas of research for both military and civilian applications. Critical to the problem is the need for fully distributed information management technologies that are efficient, adaptive and resilient. In this paper, we introduce and discuss a new strategy for tactical data dissemination and processing based on distributed online learning. Starting from a formal description of the problem we introduce our proposed solution and its theoretical properties. We also present and discuss a number of simulation experiments for different data dissemination scenarios, and conclude the work with a discussion on how such techniques may be applied to critical e-government environments under different assumptions of service availability and information release policies.
Chapter Preview

Background And Previous Work

The resource allocation problem for data processing in mobile ad hoc networks can be generally classified into three main groups: (1) local data processing, (2) remote data processing, and (3) distributed (or in-stream) data processing. In each case, the goal is to allocate resources for data processing and distribution from a source node to multiple sink nodes requiring (possibly) different variations of the data.

In the first type of problems (i.e., local data processing), the source of the data is responsible for providing the necessary transformations required by each client. Similar to conventional client-server models, local data processing essentially allocates all processing to the data source (i.e., the server). The research focus on these types of problems is basically in the allocation of resources for data distribution (i.e., data routing).

Curran (2003) proposed a reinforcement learning-based algorithm for routing in ad hoc networks. The SWARM protocol is data agnostic, focused only on packet routing. When receiving a data packet, each node chooses the appropriate action (next hop) based on current policies. The work was later extended by Dowling, Curran, Cunningham, and Cahill (2004) who proposed the collaborative reinforcement learning-based routing protocol called SAMPLE, for mobile ad hoc networks. Chang, Ho, and Kaelbling (2004) have also proposed the use of reinforcement learning techniques for data routing in mobile ad hoc networks. Although the approach did not address tactical issues such as service decomposition and distribution, it did allow for interaction between data routing and node mobility. Peng and Deyun (2006) also leverage from reinforcement learning algorithms to improve quality of service (QoS) routing strategies. In his work, Peng proposes a heuristic-based algorithm that utilizes reinforcement learning to estimate best QoS routing paths from previous experience, reducing the number of QoS flood and probing packets for path maintenance in mobile networks.

Complete Chapter List

Search this Book: