ToNC: Summary of Algorithmic Foundations Working Group

1
ToNC Summary of Algorithmic Foundations Working
Group
John Byers, Neha Davé, Joe Hellerstein, Richard
Karp, Richard Ladner, Gregory Malewicz, Satish
Rao, William Steiger, and George Varghese March
17, 2006
2
Distinctive Network Attributes and Considerations

• Massive scale and scale-invariance
• Constraints space, memory, power, processing
• Streamed data continuous queries
• Heterogeneity of capabilities
• Graphs locally known, imperfectly known, or
  hidden.
• Evolving topology
• Networks increasingly under attack

3
Algorithms Within The Network Providing
Fundamental Functionality

• Routing
• Exploiting geometry, e.g. curveball routing
• Quantifying expressiveness vs. complexity
  tradeoffs
• Avoiding/mitigating route flapping and
  oscillations
• Approximately optimal routing compact routing.
• Load balancing and scheduling
• Impact of heterogeneity complex failure modes
• Economic considerations imperfect information
• Naming and lookup
• Data-centric lookup
• Intentional naming
• Specification and validation
• Rich routing semantics that are verifiable

4
Algorithms Within the Network Making the Network
Better

• Measurement and management
• Streaming algorithms
• Going beyond AMS and FM sketches
• Network self-analysis and correction (more next)
• Fault diagnosis
• Why button, detection of correlated failures
• Network coding
• Improving defenses, detection and forensics
• Algorithmic detection of outliers or patterns
• Construction of defenses that are hard to learn
• DDoS traceback, worm propagation traceback

5
Network Self-Analysis

• Given a time-evolving graph where
• Edge deletions and insertions are frequent
• Data arrives online, one-pass access
• Graph size may be prohibitive to store in its
  entirety
• Goal 1 Compute summaries/sketches of key graph
  properties (conductance, bad cuts, eigenvalues).
• Goal 2 Have the network take corrective action.

6
Networks as Objects

• Holistic approach operate on the entire network
• Universality, simulations, and embeddings
• Can GENI simulate an arbitrary network? (more
  next)
• Network growth and dynamics
• Model, measure, exploit!
• Codesign of network and algorithms
• Networks within networks
• Overlays, underlays, Grid, P2P

7
Universal Networks

• Can GENI simulate an arbitrary network with
  different naming and routing conventions?
• Can we embed a complex application or experiment
  into a target infrastructure? Problem sketch
• Multi-commodity flow problem
• Known traffic matrix
• Routes between ingress and egress nodes known
• Goal embed this application into the
  infrastructure.
• Minimize consumed resources, interference.
• Connects to key systems issues of virtualization,
  emulation, repeatability of experimentation.

8
Lessons to Apply Going Forward

• Value of simple stripped-down models.
• Avoid pernicious effects of the ns simulator
• Lack of historical data has hindered validation
• Predict what we will need from GENI.
• Make sure to demand it, collect it!
• Diversity of processing elements communication
  media
• Indispensability of networks in society