Design and Operation of Efficient Self-Organizing Systems

At the beginning of this 21st century, companies are faced with two key challenges with regard to their IT systems: first, the increasing complexity of IT systems gives rise to escalating operational expenditures for their administration and maintenance. Second, the intensifying, worldwide competition in all markets requires IT systems providing more agility, flexibility, scalability, robustness, and adaptivity in tackling daily businesses. Consequently, companies call for IT solutions with a high degree of autonomy - in order to manage themselves - as well as a high degree of decentralization - in order to provide the required, beneficial properties.

Self-organizing emergent systems are generally acknowledged as a potential solution able to cover both of these requirements. They consist of many, simple elements (e.g. agents, servers, mobile devices, or robots), which have only partial or even no global system knowledge and make their decisions solely based on locally available information. The global coherent system behavior is achieved only by means of the local actions and interactions between the elements, each unaware of the system's goals. The problem-solving power of a self-organizing emergent system hence mainly resides in the interactions between its elements instead of the internal reasoning of individual elements.

However, there exist several problems and challenges, which hinder the acceptance of self-organizing emergent systems by industry. This thesis tackles two of them: first, the design of efficient self-organizing emergent systems today is too complex, time-consuming, and costly. Second, an acceptable efficiency of self-organizing emergent systems during the operation cannot be guaranteed for all situations. These problems form two challenging paradoxes: first, in order to conquer system complexity, one has to create more complex systems in a much more complex way. Second, in order to lower operational expenditures, one has to use potentially inefficient systems that actually may increase operational expenditures.

Thus, this thesis presents several artifacts that on the one hand simplify the design of effective as well as efficient self-organizing emergent systems and on the other hand facilitate their efficient operation even in unforeseeable situations. In more detail, the major contributions of this thesis are as follows:

1. We investigate the general principles behind decentralized coordination by means of chemical stimuli (infochemicals) between organisms in biology and adopt them in a computational coordination model. Because infochemical-based coordination (IBC) is the most universally employed communication and coordination model between organisms in biology with a plethora of inspiring examples, the formally adopted principles provide the foundation for the future specification of various, bio-inspired, decentralized coordination mechanisms using digital infochemicals. The expressiveness of the adopted model and the use of infochemicals with different semantics, dynamics, and functions, allow for the simplified design of more efficient solutions and solution processes compared to existing approaches.
2. We present a corresponding design pattern that encapsulates the adopted coordination model in a systematic way familiar to software engineers. The pattern hides the inherent complexity of the designed system and makes meaningful abstractions from the biological principles. Furthermore, we present design guidelines that support the identification and adaptation of new coordination mechanisms based on IBC. This simplifies the design of new solutions but does not force engineers to be biological experts at the same time. Moreover, we develop an adequate tool for the simulation of various coordination models and mechanisms in different application domains.
3. We present the general model of an advisor that is able to improve the efficiency of self-organizing emergent systems solving dynamic optimization problems with recurring tasks. This so-called Efficiency Improvement Advisor (EIA) realizes an unobtrusive feedback and learning mechanism that is independent of the coordination model or mechanism used by the underlying self-organizing emergent system as well as the problem domain in hand. The EIA in particular takes into account the low observability and poor controllability of self-organizing emergent systems during operation, considers their openness and basic autonomy, and preserves their beneficial self-organizing emergent properties.
4. We develop a decentralized coordination mechanism based on IBC, which takes its inspiration from the pollination of flowers by honey bees. This so-called pollination-inspired coordination (PIC) mechanism demonstrates the many beneficial capabilities of IBC and can \eg be used for the self-organizing emergent solution to every-day problems in logistics, more specifically pickup and delivery problems. Likewise we develop an instantiation of the EIA model for this mechanism and domain. An experimental evaluation proves the efficiency of the PIC mechanism as well as the achieved improvements by the EIA.

URN: urn:nbn:de:bvb:384-opus-17490
URL: http://www.opus-bayern.de/uni-augsburg/volltexte/2011/1749/