Research Challenges of Autonomic Computing

1
Research Challenges of Autonomic Computing

• Based on Kepharts paper, ICES 2005, May 15-21,
  2005

2
Related work

• IBMs autonomic computing initiative
• HPs Adaptive Enterprise initiative
• Microsofts Dynamic System Initiative

3
Framework

• Basic parts Autonomic elements, autonomic
  systems, human-computer interactions
• Autonomic elements any type of computing
  resource e.g., a storage device, a database, an
  application server, etc.---improving
  self-managing capability

4
Specific Autonomic Elements - Challenges

• Self managing components
• Effective interoperation among autonomic elements
• Components must adopt standard interfaces
• Generate and supply needed information to other
  components (what if scenario questions)
• Example An application server being asked to
  estimate the benefit of adding an additional
  server

5
Generic Autonomic Element Technologies

• Key Monitoring, event correlation, rule
  execution, modeling, optimization, forecasting,
  planning, feedback control, and machine learning
• Consistent common format for monitored events and
  log file
• Rule engines and correlation engines to analyze
  monitored data
• Extracting correlations between low-level system
  measurements and high-level services
• Need models that map potential actions into
  probable outcomes
• Models must be self-adaptive to changes
• Exploration vs. exploitation
• Planning algorithms

6
Generic autonomic element architectures, tools,
and prototypes

• Supporting multiple threads within a component
• Detecting and resolving conflicts that are bound
  to occur when an element cannot meet all its
  obligations

7
Autonomic System Challenges

• Technologies For system level self-configuration,
  self-healing, self-optimization, and
  self-protection
• Self-configuration When a new element is added,
  it should integrate itself into the system (e.g.,
  register, etc.)
• Autonomic deployment of a large-scale application

8
Human-computer Challemges

• Interfacing with humans---iterative approach to
  setting policies etc.
• Goal-based and utility-function-based policies
  not just low-level policies