Autonomic Computing: An Overview Manish Parashar and Salim Hariri

1. Autonomic Computing: An OverviewManish Parashar and Salim Hariri Presenter: Alejandro Simon Agnostic: Joseph Cilli

2. Introduction • Advances in technology have resulted in complex, heterogeneous and dynamic applications and systems. • Growth in the information infrastructure aggregates such systems. • Applications, programming environments and information infrastructures have become brittle, unmanageable, insecure. • Autonomic computing is an alternate paradigm that deals with this issue.

3. The Autonomic Nervous System • The most sophisticated example of autonomic behavior. • Regulates and maintains homeostasis: maintains structure and functions by means of a multiplicity of dynamic equilibriums that are rigorously controlled by interdependent regulation mechanisms. • Not all parameters have the same urgency, essential parameters are monitored more closely.

4. Ashby’s Ultrastable System • In order for an organism to survive, its essential variables (EVx) must be kept within a viability zone. Source: “Autonomic Computing: An Overview, ” M. Parashar, and S. Hariri, UPP 2004, Mont Saint-Michel, France, Editors: J.-P. Banâtre et al. LNCS, Springer Verlag, Vol. 3566, pp. 247 – 259, 2005.

5. Ashby’s Ultrastable System • The goal of an adaptive behavior is to ensure survivability of the system. • If external or internal disturbances push the system outside its equilibrium state, the system will work towards returning to equilibrium. • The ultrastable system consists of two closed loops, one controls small disturbances, the other controls larger ones.

6. Ashby’s Ultrastable System Source: “Autonomic Computing: An Overview, ” M. Parashar, and S. Hariri, UPP 2004, Mont Saint-Michel, France, Editors: J.-P. Banâtre et al. LNCS, Springer Verlag, Vol. 3566, pp. 247 – 259, 2005.

7. The Nervous System as a Subsystem of Ashby’s Ultrastable System • Nervous System consists of Peripheral Nervous System (PNS), Central Nervous System (CNS) • PNS: Sensory neurons connecting stimuli receptors to CNS. Motor neurons connecting CNS to muscles and glands. • CNS: Sensory-somatic nervous system and Autonomic Nervous System: Depicted as an Ashby’s Ultrastable System

8. The Nervous System as a Subsystem of Ashby’s Ultrastable System Source: “Autonomic Computing: An Overview, ” M. Parashar, and S. Hariri, UPP 2004, Mont Saint-Michel, France, Editors: J.-P. Banâtre et al. LNCS, Springer Verlag, Vol. 3566, pp. 247 – 259, 2005.

9. The Autonomic Computing Paradigm • Must have a mechanism to adapt to changes in its Essential Variables (EVs) by changing its behavior to restore equilibrium. • Equilibrium is impacted by internal environment (excessive CPU usage), and external environment (external attack). • Requires sensor and motor channels to sense and react to changes in environment by changing the system and maintaining equilibrium. • Stages: Sensing, Analyzing, Planning, Knowledge, and Execution.

10. A Holistic View of Autonomic Computing • Existing systems have been developed in an ad-hoc manner and can optimize a few attributes or functionalities. • Emerging systems and applications are dynamic, their requirements will change during their lifetime (high performance, fault tolerance, security, availability, configurability) • Autonomic computing provides a holistic approach to design and development of application that can adapt to their requirements without manual intervention.

11. Architecture of an Autonomic Element • The smallest unit of an autonomic application. • It is a self-contained software or system module with input and output interfaces and explicit context dependencies. • It has embedded mechanisms for self-management.

12. Architecture of an Autonomic Element Source: “Autonomic Computing: An Overview, ” M. Parashar, and S. Hariri, UPP 2004, Mont Saint-Michel, France, Editors: J.-P. Banâtre et al. LNCS, Springer Verlag, Vol. 3566, pp. 247 – 259, 2005.

13. Autonomic Computing Systems and Applications • Self Aware: knows itself and is aware of its state and behaviors. • Self Configuring: configures and reconfigures itself under varying and unpredictable conditions • Self Healing: detects and recovers from problems. • Self Protecting: detects and protects from internal and external attacks to its resources. • Context Aware: is aware of its execution environment and reacts to changes in such. • Open: must function in an heterogeneous world and be portable across platforms. • Anticipatory: anticipate its needs and behavior and those of its context; should manage itself proactively.

14. Autonomic Computing Research Issues and Challenges • Conceptual: Defining models for controlling and implementing autonomic behavior. • Architecture: Implementation of robust and predictable autonomic behaviors. • Middleware: Implement reliable and robust middleware to allow autonomic entities to discover, message, and trust each other. • Application: Creation of frameworks to allow systems and applications that are capable of managing themselves.

15. The Autonomic Computing Landscape • Academic and Industry projects have addressed and investigated the issues outlined above. • A list follows of applications that utilize autonomic mechanisms for problem determination, monitoring, analysis and management.

16. The Autonomic Computing Landscape Source: “Autonomic Computing: An Overview, ” M. Parashar, and S. Hariri, UPP 2004, Mont Saint-Michel, France, Editors: J.-P. Banâtre et al. LNCS, Springer Verlag, Vol. 3566, pp. 247 – 259, 2005.

17. Conclusions • Autonomic Computing Paradigm: inspired by biological systems such as the human nervous system. • This paradigm enables the development of self managing computing systems and applications. • Autonomic strategies and algorithms handle complexities and uncertainties with minimum human intervention. • Research is being conducted but this topic remains an open and significant challenge.

18. Agnostic Question 1 • Being that the ACS paradigm is modeled after the human nervous system, is it accurate to conclude that the paradigm is fallible when you consider that a system might not be able to self-heal an unknown/unforeseen issue just as the human biological system may not be able to self-heal because it is unaware of specific infections and its inability to learn unless combined with artificial intelligence? • I think it is accurate to say that the self-healing capabilities of an ACS depend on the sophistication of the algorithm. Even with the use of artificial intelligence, the possibility of the system being exposed to an unforeseen issue will always exist. Therefore, consulting an expert is a wise course of action.

19. Agnostic Question 2 • The authors’ state an ACS requires sensor channels to sense changes and motor channels to react; however, there is never any mention of the overhead involved with sensing, analysis or reaction. Is there a significant cost to these requirements that is overlooked in the paper? • There is always computational overhead involved in executing extra code. However, use of threads and performance analysis of the application to ensure it meets the contractual Quality of Service should ensure that the overhead introduced by the monitoring and healing code does not appear to hinder the application. There is a substantial cost in terms of programming time when implementing these requirements.

20. Agnostic Question 3 • The idea of ACS seems like a viable solution to handle complex systems. Yet there have been instances where an ACS has caused more work for an administrator. Have there been any studies to compare system performance using solely human interaction with system performance using an ACS? • Not familiar with a specific study comparing the performance of human interaction vs. an ACS.

21. Agnostic Question 4 • Once high level policies are defined by a human, how would an ACS handle conflict resolution between contradictory self management aspects? • Several algorithms can be implemented to implement conflict resolution. The most trivial one is to require human input. Another options is to implement either one of the conflicting strategies, monitor and record its effect in the system. After several iterations, choose the one with the highest success rate. This approach would mimic the way humans learn by making choices with incomplete information and learning from their mistakes.

22. Agnostic Question 5 • The authors’ state that one of the research challenges is autonomic application & system architecture. Since the proposed architectures deal with autonomic element communication, can one infer that such architectures would be OS and/or machine specific? • Not necessarily, with the advert of managed languages and open communication standards such as Java, .NET languages, HTTP and SOAP, such architectures need not be OS nor machine specific.

23. Agnostic Question 6 • Once a human sets high level policies, how can a non-programming administrator determine if the self-management aspects of a system are performing properly/efficiently? • A non-programming administrator is limited to monitoring the system as a black box. Such monitoring can include query execution times, and other statistic reported by the self-management modules.

24. Agnostic Question 7 • Clearly ACS has a broad use in many industries; however, the authors’ examples of existing projects mostly focus on data management systems. Was the current concept of ACS spawned by DBMS or does it simply lend itself to that arena? • Each author would choose examples that are most familiar to them. The concept of ACS applies to DBMS given the complexity of such systems and the familiarity of Computer Science majors with such systems. ACS however, are not limited to DBMS as the examples below show: • “An Al tool for supervising substations”, Melvin Ayala, S. Galdenoro Botura, J. Oscar Maldonado, IEEE POTENTIALS, 2002, VOL 20; PART 5, pages 13-18 • “http://www.ForexLab.NET”, A mechanical trading software.

25. Agnostic Question 8 • The authors’ conclude that achieving overall autonomic behaviors remain an open and significant challenge, which can be accomplished, in part, by open industry standards; however, a lone standards organization body has yet to be defined. How can the stated challenges be met when multiple governing bodies exist? • The same way problems have been solved in the past, several groups implement a solution to the problem, a particular implementation becomes a de-facto standard and the market determines which solution fits each particular niche.