A Scalable Architecture for Network Fault Diagnosis

1
A Scalable Architecture for Network Fault
Diagnosis

• George Lee
• Steven Bauer
• Peyman Faratin
• Advanced Network Architecture Group
• 9/19/2005

2
Network fault diagnosis is hard
Jane cannot connect to Yahoo. What should she do?
Internet
Yahoo!
Jane
In the Internet today, there is no easy way to
find out what is happening in the network
3
The Knowledge Plane allows agents to exchange
diagnostic data
Knowledge Plane
ISP 1 Diagnostic Agent
ISP 2 Diagnostic Agent
ISP 3 Diagnostic Agent
Yahoo!
Jane
Diagnostic Agents exchange failure data, perform
diagnosis, and tell the user who is responsible
for their problem.
4
Architectural research challenges
Knowledge Plane
ISP 1 Diagnostic Agent
ISP 2 Diagnostic Agent
ISP 3 Diagnostic Agent
Representation
?
Routing
Scalability
Bob
Yahoo!
Jane
Chris
5
ISP-based diagnosis
Knowledge Plane
ISP 1 Diagnostic Agent
ISP 2 Diagnostic Agent
ISP 3 Diagnostic Agent
Jane
Yahoo!
ISP 1
ISP 2
ISP 3
6
Aggregation can improve scalability
Knowledge Plane
ISP 1 Diagnostic Agent
ISP 2 Diagnostic Agent
ISP 3 Diagnostic Agent
Jane
Yahoo!
ISP 1
ISP 2
ISP 3
7
Agents aggregate requests using a network
dependency model
ISP 1 DA
ISP 2 DA
Yahoo!
Jane
ISP 1
ISP 2
ISP 1 DAs Network Model
Jane connects to Yahoo!
Tasks
depends on
ISP 2
Janes LAN
ISP 1
Yahoo!
Components
8
(No Transcript)
9
Conclusion and Next Steps

• Conclusion
• Dependency-based aggregation can greatly improve
  the scalability of distributed diagnosis.
• Next Steps
• Evaluate aggregation for a wider variety of
  failures.
• Build a prototype diagnostic architecture and
  prototype agents.