Junichi Suzuki, "Biologically-inspired Autonomous Adaptability in Communication Endsystem: An Approach Using an Artificial Immune Network," Ph.D. Dissertation, Keio University, 2001.
Abstract:
The explosive growth of the net places larger and more challenging demands on underlying communication system. However, the design, optimization and deployment of communication system are still hard while computing power and network bandwidth have increased dramatically. It has a remarkably rich set of options, or policies, and no single policy fits all applications or environments. Therefore, it is essential to develop an open-ended and adaptive framework that allows to tailor its applications by configuring their optimal settings out of feasible policies. It is tedious, error-prone and economically expensive, however, to understand the tradeoffs among alternative policies and navigate through them manually. Developers face the chaotic efforts of engineering a system, resulting in ad-hoc solutions. Such systems are often hard to maintain, customize and tune, since much of the engineering tasks are spent just for trying to get the system operational. Moreover, the cost of reconfiguring communication system components is often not linear; small changes can result in large effects, and large changes in small effects. The next generation communication system should address two major issues of today; system adaptability and autonomous system adaptability. This study proposes a solution to these challenges by (1) designing communication system based on reflective meta-architecture, and (2) adopting the biological concepts and mechanisms in order to give autonomy to the system's behavior.
This study presents OpenWebServer/iNexus, which is a research vehicle for investigating and demonstrating autonomous adaptability in communication endsystem. OpenWebServer is both an adaptive web server and an object-oriented framework that allows to tailor optimally configured network services. Designed based on a reflective meta-architecture, it contains a metalevel that abstracts a wide range of aspects, i.e. structures and/or behaviors, of web servers. They include concurrency, I/O event dispatching, protocol parsing, connection management, caching, logging, service redundancy. The modular architecture of OpenWebServer allows maintaining these aspects and their policies explicitly as fine-grained metaobjects. The system structure and behavior can be altered statically and dynamically by adding, deleting, changing or replacing metaobjects.
iNexus is a policy coordination engine embedded within OpenWebServer. It serves an automated policy-based management of web services and applications. Keeping complex adaptive system like biological systems in mind, it is designed as a self-organising facility to autonomously reconfigure system's internals according to requirement changes in applications and environmental changes. The structure and behavior of iNexus are modeled with the principles and mechanisms in the natural immune system, because its behaviors, e.g. self-protection and self-regulation, are perfect examples of system's autonomous adaptability. iNexus artificial immune network examines OpenWebServer's performance and resource usage periodically, measures a delivered quality of service, and selects a suitable set of policies to reconfigure the system dynamically. iNexus coordinates a series of policies by relaxing constraints among them. The policy coordination process is performed through decentralized interactions among policies without a central point of control, as the natural immune system does.
In order to evaluate adaptability in OpenWebServer/iNexus, this study conducted three kinds of experiments: (1) scalability test, (2) static adaptability test and (3) dynamic adaptability test. The first experiment conducted a systematic benchmarking of different configurations of OpenWebServer/iNexus under different server load conditions. The result showed how flexible nature of OpenWebServer/iNexus allows adapting from its baseline performance to stable high-throughput performance. It demonstrates it is possible to improve server performance through superior server design. A flexible framework like OpenWebServer/iNexus need necessarily not perform poorly, while a hard-coded server can provide excellent performance. The second experiment examined static adaptability of OpenWebServer/iNexus by observing which policies iNexus selects according to the underlying platform's features (e.g. the number of CPUs, the types of supported I/O functions, and thread support), when the system starts running. Through several test cases, iNexus demonstrated it selected reasonable set of policies statically at the system's startup time. The third experiment examined autonomous adaptability of OpenWebServer/iNexus by observing how iNexus changes policies dynamically at runtime. It runs the system in the situation where load condition bursts. In this test run, iNexus allowed OpenWebServer to dynamically reconfigure its internal policy combination so that it increases its throughput yet keeps low average error rate. The result showed OpenWebServer/iNexus adapts to environmental changes in the peacemeal way by altering itself.
OpenWebServer/iNexus has been used as a foundation component in a distributed collaborative environment, called xSpaces, to demonstrate its feasibility in real-world applications. xSpaces is designed to support applications for distance learning, distance conferencing or distributed collaborative work. It has been used for constructing a self-paced distance learning application and group-paced collaborative space. These applications are intended to be used by universities providing online courses for their students or distance learners, and private organizations that wish to keep their workforces well trained. The target educational domain is software modeling. The self-paced learning courses are used for teaching general software modeling techniques for novice learners. The group-paced training courses are used in a team setting, where participants collaboratively model software structure and behavior. Through the experience of developing these two applications, OpenWebServer/iNexus demonstrates that it is fairly robust, adaptable and scalable.
This study describes a biologically-inspired approach to design communication system that dynamically adapts to changing environments. OpenWebServer/iNexus supports autonomous policy coordination and system reconfiguration as built-in mechanisms. Augmented by an artificial immune network, it can select a suitable set of policies through decentralized interactions among them. This study illustrates that adaptability and scalability of communication system can be leveraged by producing a synergy effect of reflective meta-architecture and artificial immune network. It provides a blue print showing autonomous adaptability in next generation communication system.
Acknowledgements:
I would like to thank all the people who made this work possible by helping shape my research vision, sharing insights, discussing ideas, reviewing papers, and encouraging me.
I wish to express my sincere gratitude to my advisor, Prof. Yoshikazu Yamamoto, for his invaluable guidance, support and patience during my Ph.D. program. He was always open to discussions, and gave me the freedom and means to mature my ideas while always pointing me the right direction. It has been extremely rewarding to have him as my advisor.
My gratitude also goes to Prof. Norihisa Doi, Prof. Yuichiro Anzai, and Prof. Masahumi Hagiwara for serving on my dissertation committee. They invested the time to give me feedbacks for earlier drafts of this dissertation through their careful readings.
Throughout this doctoral study, I have collaborated with many brilliant people who improved my work. Many discussions with Toshimitsu Minami and Mio Yamamoto about artificial immune system made my rough research vision more real and feasible. They contributed to develop iNet, a framework for building artificial immune networks (see Section~\ref{chap:inexus}), and expanded its use cases for the pursuit game, a standard problem in the field of multi-agent system, and a Robocup soccer game simulation. After this initial success, Yutaka Abe, Hisayo Koyabu, Naotaka Tanaka, Shintaro Tanaka and Masamichi Yamaguchi joined our Biologically-inspired Software Architecture group. This working group has been highly constructive so that we can share insights with each other. I gratefully acknowledge Yutaka for our earnest and stimulating discussions on applying pheromone emission into his information retrieval application. Also, I would like to thank to Tatsuya Suda, Tomoko Itao, Masato Matsuo, Tetsuya Nakamura, Miyuki Imada, Naomi Miyamoto and Akihiro Enomoto. Through many discussions, they have always gave me generous inspirations and insightful ideas to design the communication system inspired by biology.
For the contribution to the UXF (UML eXchange Format) and xSpaces projects (see Chapter 6), I would like to thank to their supporters: Daniel Amyot, Prof. Jean Bezivin, Peter van Eijk, Lars Marius Garshol, Andrey Khavryutchenko, Gordon Miller, Uche Ogbuji, Wei-Guo Peng, Ivan Porres Paltor, Roland Rosenfield, Kenji Shirane and Yun Tanach. I would also like to thank to Yutaka Abe and Kei Fuji for helping implement the Persona toolkit (see Chapter 6) to develop web content personalization engine. Gaku Tashiro and Shogo Tsuji helped to make Persona's idea and implementation mature.
I would like to express my sincere appreciation to the people who I have worked with at my employers during my Ph.D. project, Soken Planning Co., Ltd., Object Management Group Inc., Object Management Group Japan, Inc., and Institute for Social Engineering: Sachiko Aoki, Tadaaki Arimura, Suwako Doi, Yoko Furuhashi, Satoka Hasegawa, Yoko Hirano, Ken Ichida (currently at Dentsu, Inc.), Yurie Ito, Hiroki Kamata, Eriko Saito, Megumi Senda, Hiromu Shibata, Toshikazu Shinohara, Jon Siegel, Richard Mark Soley, Tomoo Takeda, Mari Uchida, and Hiroshi Wada. They provided me a highly stimulating and warm working environment. Its atmosphere made my last eight years a continuous and intense learning experience.
It is not possible to enumerate everybody who supported me, but in particular, I wish to recognize Masatoshi Iwano, Ed Fernandez, Hiroshi Hayata, Masaru Ishii, Hideki Kaneko, Naoto Kawasaki, Robb Keayes, Linda Rising, Koichi Magai, Katsuo Makino (in memory), Ryo Matsuda, Nozomu Matsui, Yoshiro Mihira, Daniel Ming-Tung, Osamu Nakamura, Wataru Nyui (in memory), Ryuichi Nakase, Keiko Nakase, Yuuya Sugita, and Yun Tanach.
Special thanks go to Hiroki Kamata. For eight years since we met for the first time, he has sparked my interests in various domains such as literature, psychology, philosophy and biology as well as computer science, which undoubtedly helped me to complete my degree. His profound understanding and perspective to science, engineering and human nature makes him a unique master, which will be the standard that I will always strive to.
My parents have given me unrelenting support for longer than I can remember. This is the first chance to thank them publicly, in writing. Recounting the reasons this work would not have been possible without them would require an another tome. I love them.
The largest share of the credit goes to Kumiko, my wife and great partner. She has tolerated my preoccupation with the Ph.D. work, and constantly kept my spirit up through three years of painstaking research and writing. Her unfailing support and love have been pivotal for the successful completion of a long journey.
Junichi Suzuki
March, 2001
Yokohama City, Japan and then Irvine CA, U.S.A.