Biologically-inspired self-adaptation mechanisms for network applications

Network applications are increasingly required to be autonomous, scalable, and adaptive to dynamic changes in the network. Based on the observation that various biological systems have already satisfied these requirements, this paper proposes and evaluates a biologically-inspired adaptation framework that makes network applications to be autonomous, scalable and adaptive. With the proposed framework, called iNet, each network application is designed as a decentralized group of software agents, analogous to a bee colony (an application) consisting of multiple bees (agents). Each agent provides a particular functionality of a network application, and implements biological behaviors such as migration, reproduction, energy exchange and death. iNet implements adaptive behavior selection mechanisms for agents. iNet is designed after the mechanisms behind how the immune system detects antigens (e.g., viruses) and produces specific antibodies to eliminate them. It models a set of environment conditions (e.g., network traffic and resource availability) as an antigen and an agent behavior (e.g., migration) as an antibody. iNet allows each agent to autonomously sense its surrounding environment conditions (an antigen) to evaluate whether it adapts well to the sensed conditions, and if it does not, adaptively perform a behavior (an antibody) suitable for the environment conditions.This dissertation research studies the optimality and stability of adaptation in iNet. Corresponding to those properties, two adaptation mechanisms are proposed: iNet-GA and iNet-EGT. iNet-GA allows agents to seek their optimal configurations that optimize their performance (e.g., response time). In iNet-GA, a configuration of antibodies is encoded as a set of genes, and antibodies evolve via genetic operations such as crossover and mutation. iNet-EGT allows agents to seek their stable configurations that reduce the fluctuations in their performance. In iNet-EGT, the behavior selection process is modeled as a series of evolutionary games among behaviors. It is theoretically proved to converge to an evolutionarily stable (ES) equilibrium a specific (i.e., ES) behavior is always selected as the most rational behavior against a particular set of network conditions. This means that iNet-EGT allows every agent to always perform behaviors in a rational and adaptive manner. Simulation results show that agents adapt to dynamic and heterogeneous network environments by evolving their antibodies across generations. The results also show that iNet allows agents to scale to workload volume and network size.