Coordination Strategies: Classifications

Let's take a look at the ways one can slice up the problem of coordinating agents that deal with multiple users. Here's a partial list of ways one might want to slice things up. I haven't rigorously defined all of these terms---they seem fairly intuitive---but anyone who wants tighter definitions should let me know and I'll add them to this list.

Degree of decentralization of the agent
Domain characteristics of the agents
- Speed of response
  - Subsecond
  - Subminute
  - Daily
  - Longer
- Robustness & accuracy required
  - Life-critical
  - Enterprise-critical
  - Advisory
  - Informational
  - Fun
  - Deliberately misleading
- Security of information
  - Stored in agent
  - Between coordinating servers or peers
  - Transmitted to user
  - Threat model
    - Denial-of-service
    - User impersonation
    - Exposure of confidential information
    - Modification of information
    - Perversion
- Source of domain knowledge
  - External databases
  - Users (excluding simple personalization inputs)
- Completeness of domain knowledge
  - Complete
  - Probably complete
  - Known partial
- Persistence of agent memory
  - Long-term/permanent
  - Per-user
  - Per-transaction
  - None
- User interface issues
  - Degree of anthropomorphism
  - Exposure of underlying mechanism (e.g., other agents)
Coordination model
Communications infrastructure employed
- UDP or IP connectionless
- TCP connection-oriented
  - Store-and-forward (mail)
  - Real-time (telnet, web)
- Multicast
  - Low bandwidth
  - High bandwidth (e.g., MBONE)

Previous | Table of contents | Next

Lenny Foner

Last modified: Fri Dec 15 10:36:00 1995