A MultiAgent Systems Approach to Autonomic Computing

1
A Multi-Agent Systems Approach to Autonomic
Computing

• Geradl Tesauro, David M. Chess, William E. Walsh,
  Rajarshi Das, and others
• IBM T.J. Watson Research Center

Mingliang Wang
2
MAS vs. AC
Application, adaptable for
A system composed of several agents, capable of
reaching goals that are difficult to achieve by
an individual system
Self-Configuration, self-Healing,
self-Optimization, self-Protection
Flexible, high-level interactions
Automatic group formation, emergent behavior,
multi-agent adaptation, agent coordination
Spur basic research
3
UNITY Structure
Application Manager
Resource Arbiter
Registry
Policy Repository
sentinel
4
UNITY Structure cond
Registry

• High level policy
• Utility function

App Manager
User Interface
Query
.
R. Arbiter
Performance, states of elements
5
Self-Assembly
Contact info for registry
High level goals, directions
Policy Repository
Other AEs
AE

• Contact Registry
• Find other elements
• Manage Relations
• Register self

6
Self-Healing of Policy Rep.
Resource Arbiter
2. Start another instance
1. Sentinel detects failure
Joining existing cluster of synchronized policy
repositories Replicating data including
subscription themselves
Policy Rep.
Sentinel
Policy Rep.
Policy Rep.
7
Data Center Self-Optimization
Business value F ( service level rendered )
Embedded in application manager
8
Data Center Self-Optimization contd
9
Discussions

• Where are the multiple agents? Compared to with
  previous architecture.
• How are the applications scheduled in different
  machines? IBMs Topology Aware Grid Service
  Scheduler
• Resource Virtualization of Servers
• Whats the cost to add this MAS-AC to the
  existing applications?