Research Challenges of Autonomic ComputingJeffrey O. KephartIBM Thomas J. Watson Research Center19 Skyline Drive, Hawthorne, NY 10532, [email protected] computing is a grand-challenge vision of the future inwhich computing systems will manage themselves in accordancewith high-level objectives specified by humans. The IT industryrecognizes that meeting this challenge is imperative; otherwise, ITsystems will soon become virtually impossible to administer. Butmeeting this challenge is also extremely difficult, and will requirea worldwide collaboration among the best minds of academia andindustry. In the hope of motivating researchers in relevant areasto apply their expertise to this vitally important problem, I outlinesome of the main scientific and engineering challenges that collec-tively make up the grand challenge of autonomic computing, andprovide pointers to initial efforts to address these challenges.Categories and Subject DescriptorsA.1 [GeneralLiterature]: Introduction and Survey; H.0 [InformationSystems]: General; H.4 [Information Systems Applications]: Gen-eral; I.O [Computing Methodologies]: General; I.2.11 [DistributedArtificial Intelligence]: [intelligent agents, multiagent systems];K.6 [Managementof Computing and Information Systems]: Gen-eralGeneral TermsManagement, Measurement, Performance, ReliabilityKeywordsAutonomic computing, self-managing systems, research challenges1. INTRODUCTIONTypical present-day IT1environments are complex, heteroge-neous tangles of hardware, middleware and software from multiplevendors that are becoming increasingly difficult to integrate, install,configure, tune, and maintain. At the present rate of growth in com-plexity, even the most skilled IT professionals may find it impossi-ble to administer IT environments within a few years. Most of the1Information TechnologyPermission to make digital or hard copies of all or part of this work forpersonal or classroom use is granted without fee provided that copies arenot made or distributed for profit or commercial advantage and that copiesbear this notice and the full citation on the first page. To copy otherwise, torepublish, to post on servers or to redistribute to lists, requires prior specificpermission and/or a fee.ICSE’05, May 15–21, 2005, St. Louis, Missouri, USA.Copyright 2005 ACM 1-58113-963-2/05/0005 ...$5.00.IT industry recognizes that the only viable solution to this loomingcrisis is to endow systems and the components that comprise themwith the ability to manage themselves in accordance with high-levelobjectives specified by humans [38]. IBM introduced this vision ofself-managing systems in 2001 when it launched the AutonomicComputing initative [24]. Hewlett-Packard’s Adaptive Enterpriseinitiative [23] and Microsoft’s Dynamic Systems initiative [44] arerelated industry efforts that recognize that self-managing compo-nents and systems are vital to the future of IT.The purpose of this paper is to decompose the grand challenge ofautonomic computing into several of its constituent scientific andengineering challenges. Even incremental steps towards meetingsome of these individual challenges are likely to be beneficial to in-dustry in the near-term, long before the vision of autonomic com-puting is realized fully. As evidenced by over 20 workshops andconferences devoted to the topic over the last two years, autonomiccomputing offers a rich application domain for researchers in manybranches of computer science, including software architecture, sys-tems, artificial intelligence, and human-computer interfaces. It isalso likely to benefit from advances in applied mathematics, novelapplications of economic mechanisms, and even ethnographic stud-ies. It is hoped that this brief survey of autonomic computing re-search challenges will encourage researchers in relevant areas tojoin this movement, and lend their expertise to problems whose so-lution would have an enormous impact on the IT industry.The paper is organized as follows. First, in section 2, I de-scribe a framework that has proved to be useful in defining anddescribing IBM’s autonomic computing research program. Then,in sections 3, 4, and 5, I use this framework to highlight a subset ofchallenges that define fruitful avenues for research—the pursuit ofwhich could yield significant practical benefits in the intermediateto long term. I close with a summary and some final comments insection 6.2. FRAMEWORKAutonomic computing draws upon an enormous diversity of fieldswithin and beyond the boundaries of traditional computer science.Therefore, any attempt to categorize and describe the whole re-search effort and its associated challenges is bound to be imperfect.The architecturally-flavored framework that I use in this paper hasproven to be helpful in defining and growing IBM’s own autonomiccomputing research program, which now comprises the diverse ef-forts of at least 100 researchers, and can thus be viewed as a micro-cosm of the broader worldwide effort on self-managing systems.At the very coarsest level, I divide the research space into threebasic parts: autonomic elements, autonomic systems, and human-computer interactions.Autonomic elements are the basic building blocks of autonomic15systems, which through their mutual interactions produce the over-all self-managing behavior of autonomic computing systems. Es-sentially any type of computing resource can be viewed as an auto-nomic element: a storage device, a database or application server,a middleware component, a load balancer, a workload manager, aresource broker, etc. One can think of these autonomic elementsas services within a service-oriented architecture. Alternatively,and in many cases more accurately, one can identify autonomicelements as software agents, and autonomic computing systems asmultiagent systems. For more detail on autonomic elements, theirbehavior, and their interrelationships, see reference [38].Within the autonomic element branch of the research framework,I distinguish three sub-branches:• Specific autonomic elements. Research directed towardsimproving the self-managing capability of specific compo-nents such as databases, storage systems, servers, etc.• Genericautonomic elementtechnologies. Research on tech-nologies that are generally applicable to autonomic elements,including planning, modeling, forecasting, optimization, etc.• Generic autonomic element architectures, tools, and pro-totypes. Research on the internal