Design for Network Managability

1
Design for Network Managability

• Mung Chiang and Jennifer Rexford
• Princeton University
• March 2007

2
Making Networks Easier to Manage

• The boundary between network elements and the
  systems that manage them
• Managing IP networks is inherently hard
• Not designed with management in mind
• Induces difficult-to-solve management problems
• Leading to expensive, convoluted solutions
• Design for manage-ability
• New architectures that are easier to manage
• Easier to control, and easier to measure

3
Managing a Network
Network-wide objectives
optimization
tomography
measurement (dials)
control (knobs)
Network
4
Our Goals and Approach

• Our philosophy
• Change the system to make it easier to manage
• Our goal understand fundamental trade-offs
• Optimization optimality vs. complexity
• Tomography accuracy vs. overhead
• To advocate judicious changes to the network
• Our approach
• Analyze limitations of the existing system
• Explore two case studies on each topic
• Identify principles for design for manage-ability

5
Design for Optimizability

• Optimization control
• Configuring tunable parameters to optimize
• E.g., traffic engineering within a single domain
• E.g., selecting egress point in interdomain
  routing
• Challenge optimal solutions infeasible/intractabl
  e
• Today TE within a single domain
• Optimal traffic engineering with link-state
  routing
• Online traffic management in future networks
• Goal understand optimality vs. complexity
• Talks by Dahai Xu and Jiayue He

6
Design for Tomography

• Tomography measurement
• Inferring properties from the measurement data
• E.g., detect and localize path-quality problems
• E.g., traffic matrix from link-load statistics
• Challenge highly under-constrained problem
• Today infer path quality despite adversaries
• Passive sampling rather than active probing
• Accounting for adversaries in the network
• Goal understand accuracy vs. overhead trade-off
• Talks by Yannis Avramopoulos Sharon Goldberg

7
Design for Network Manageability (Seedling)
J. Christopher Ramming, STO 571-218-4373
Design for Manageability
Network Management

• DESCRIPTION / OBJECTIVES / METHODS
• Reveal dimensions in which network management is
  a technical challenge rather than simply
  programming
• Perform trade-off studies concerning network
  management design dimensions
• Discover design principles for network
  manageability
• Two case studies on tomography (input to network
  management) identify architectural elements that
  induce simpler network inference problems
• Two case studies on control (output from network
  management) identify architectural elements that
  induce simpler control problems

Existing protocols and monitoring
Well-defined management tasks
Tractable tomography and control problems
Hard tomography and control problems
New and improved protocols monitoring
Ad hoc network management
SCHEDULE AND BUDGET (9 month Seedling)
MILITARY IMPACT

• Technical foundations for future network
  management software initiatives
• Reduced network-management costs
• Easier configuration troubleshooting
• Fewer man hours simpler software
• Fewer network outages
• Faster network troubleshooting
• Improved network efficiency
• Protocols with smaller optimality gap
• Protocols that are easier to tune
• Measurements that match the tasks

Case Study
1Q07
2Q07
3Q07
Link inference Traffic matrix estimation Inter-AS
routing Egress point selection
Total budget 250k