Overview and Goals

There is wide recognition among autonomic computing researchers that autonomic systems need to be highly adaptive in several respects. Today's computing systems, as well as the workflow characteristics and business processes that they support, are continually evolving, sometimes on rapid timescales. This implies that hard-coded system models or rules for systems management behavior can be expected to have only a limited useful lifespan. Several lines of research that have been disparate to date aim to tackle this problem by developing methods to enable online adaptation of autonomic models or rules as systems and process continually change. One widely used approach employs online parameter estimation methods to adaptively set parameters, say, within a queuing model framework. A second approach, often used in policy-based management systems, aims to develop methods for automatically adapting and refining rule-based policies. A third approach, based on control-theoretic formalism which aims to apply system identification and other adaptive control methods to autonomic systems, has also been demonstrated successfully. More recently, a fourth approach has been advocated that draws upon algorithms developed within the Machine Learning community for supervised, unsupervised, and reinforcement learning.

In addition to online adaptation, there is also substantial research interest in offline adaptive methods, which are often too data-intensive or compute-intensive to perform online. Examples include batch reinforcement learning and genetic algorithms. Such methods may provide an automated way to produce sophisticated models of more complex systems that would be too time-consuming to develop by explicit knowledge engineering.

The goal of this workshop is to bring together disparate communities of researchers working on various alternative approaches to adaptation in autonomic systems, thereby creatingopportunities for cross-fertilization among communities that have been addressing common problems in relative isolation from each other. We intend that the workshop shall help clarify the relative advantages and disadvantages of each approach, as well as spark ideas for hybrid approaches combining multiple methods. Additionally, we expect the workshop to generate new ideas forhow adaptive approaches may be combined effectively with explicit domain knowledge engineering.

Specifically, this workshop will aim to:

• Elucidate the relative strengths and weaknesses of each type of adaptive approach.

• Characterize which of these approaches are best suited to specific application domains.

• Characterize the properties of application domains that make them especially amenable to particular hybrid approaches.

• Characterize how each of these adaptive approaches may also be effectively hybridized with explicit domain knowledge engineering.

Format

This is a full-day workshop. There will be an opening tutorial, 4-5 invited talks and shorter contributions from researches in industry and academia as well as a 90-minute panel discussion. Each presentation will be followed by 5-10 minutes of discussion on the aspects detailed earlier in the overview. The workshop is intended to be accessible to the broader autonomic computing community and to foster communication among different fields.

Submission Instructions

We invite submissions of extended abstracts (up to 5 pages, not including bibliography) for the short contributed talks and/or posters. The submission should present a high-level description of recent or ongoing work related to the topics above. Please e-mail submissions to amacs07workshop@watson.ibm.com as attachments in Postscript or PDF, no later than April 6, 2007.

Dates & Deadlines

Paper Submission: April 6

Notice of Acceptance: April 17

Information

Workshop URL: research.ihost.com/amacs07

Submission: amacs07workshop@watson.ibm.com

ICAC 2007: www.acis.ufl.edu/~icac2007/