Knowledge Plane and Contextbased management

1
Knowledge Plane and Context-based management

• Kaisa Kettunen
• Helsinki University of Technology / S-38.4030
• Seminar 26.-29.5.2006

2
Internet today

• Internet has become a global communication
  medium. The success derives from the fundamental
  design principle
• simple and transparent core with intelligence
  at the edges
• which is behind the strength of the Internet
• generality and heterogeneity
• rich end-system functionality
• decentralized, multi-administrative structure
• but it is also responsible for the existing
  limitations
• frustrated users when something fails
• high management overhead (manual configuration,
  diagnosis, design)

3
Context-based management

• Ambition towards dynamic operating environment
  for improved and more automated management
• Contextual approach
• Collective actions to support and provide a
  desired global outcome
• This suggests a pervasive and context aware
  environment, which would allow network
  administrators to view the status and performance
  of their devices on a variety of statistics and
  thus improve planning and management of the
  network in terms of for example
• Security
• Quality of Service
• Roaming (e.g. billing and authentication)

4
Context Aware Applications

• Adapt behavior with minimum user attention based
  on available sensor information, which has been
  converted into the format and level needed by the
  application
• Emphasis on using information instead of
  obtaining it
• Decomposition of the application into entities
  providing building blocks
• Loose coupling between applications and needed
  data
• Specification of data by its properties rather
  than physical location
• Context Servers (CS) provide maintenance,
  messaging, registration, configuration and
  mobility services to Context Entities (CE) and
  Context Aware Applications (CAA) in their range
  and enable interaction towards other ranges
• CE and CAA are abstractions of a data source or
  processing component, which actively query events
  from (other) CE entities

5
Knowledge Plane (KP)

• Pervasive system within the network
• Builds and maintains information on network
  behaviour to the users, operators and to itself
• Enhances ability to manage the network
  intelligently without disturbing the control and
  data planes
• Assembly from high level instructions and
  re-assembly on changes
• Automatic problem detection and fixing with
  indication if not possible
• Cognitive system
• Learn reason to act or propose actions
  accordingly
• Ability to handle and perform with conflicting or
  wrong information or high-level goals

6
Attributes of the KP
Global perspective Information from edges
combined with data from different parts of network
Edge involvement Knowledge produced, managed
and consumed beyond traditional edge of the
network
Compositional structure Operate in presence of
imperfect information and different objectives
Cognitive framework Respond, reason, mediate and
automate to be aware
Unified approach Common standards and framework
to structure based on knowledge, not the task
7
Knowledge Plane Architecture
Knowledge Plane
assertions
Knowledge (cognitive computations)
observations
explanations
Sensor
Actuator
Internet

• Information handling and control
• Observations describe current conditions
• Assertions capture high-level goals, intentions
  and constraints on network operations
• Explanations create conclusions from observations
  and assertions
• Learning and environment altering
• Sensors are entities that produce observations
• Actuators are entities that change behavior (e.g.
  change routing tables or bring links up or down)
• Knowledge is based on cognitive computation
  realized by artificial intelligence (AI)
  algorithms

8
What is Knowledge Plane good for?

• Fault diagnosis and mitigation
• Learning combined diagnosis and mitigation with
  interaction towards the user
• Automatic (re)configuration
• Continous and recursive detection and adjustment
  of configuration to be the optimal
• Support for overlay networks
• Instead of application level probing to evaluate
  and seek better paths, use application and
  network information collected and offered
• Knowledge-enhanced intrusion detection
• Data collection and gathering basis for next
  generation tools with several observation points

9
Sophia Knowledge Plane incarnation

• Distributed system deployed on PlanetLab that
  stores, propagates, aggregates and reacts to
  observations on network conditions without the
  learning aspect of Knowledge Plane.
• System optimizing its performance on caching,
  evaluation scheduling and planning
• Computational model using declarative programming
  language based on Prolog for evaluating and
  expressing application domain statements through
  logic rules, facts and expressions (instruction
  set)
• Example
• Each nodes local core implemented as loadable
  modules with
• Logic terms database which can be updated to
  extend the system
• Local unification engine based on standard logic
  unification
• I/O interfaces towards sensors and actuators
• Remote evaluator handling networking and protocol
  towards other nodes for delegating tasks
• Expression scheduling mechanism for maintaining
  calendar for future scheduled evaluations

eval(bandwidth(env(node(id42),
time(Sometime)), BwVar))
10
Examples

• Semantic-Enhanced Distribution Adaptation
  Networks (SEDAN)
• Content delivery and adaptation managed by
  maintained sematic information on content,
  infrastructure and clients
• E.g. Semantic-accurate content adaptation under
  resource constraints
• Formally defined data model used to organize and
  store information, e.g. scenes of a movie
  (content), service processing requirements
  (services), locations of network resources
  (resources) or user profiles (clients)
• Knowledge plane used for semantic information
  sharing between components
• Distributed decision making on decisions plane
  utilizing knowledge plane information
• Pricing mechanism for aggregate, user-centric
  utility maximization
• Manipulation of elastic users with pricing
  signals to gain optimal network resource usage
  (e.g. bandwidth or routing)

11
Examples (2)

• Protection routing algorithms on optical (GMPLS
  over WDM) networks
• Enhance network reliability, e.g. link failure
  probabilities, and thus total bandwidth
  consumption as well as decrease packet loss
• Abnormalties in link behaviour are detected based
  on learned link patterns and the information used
  to select right links or backup paths with faster
  routing algorithm computation
• Self-Management in Chaotic Wireless Deployments
• Chaotic (unplanned and unmanaged) wireless
  networks may be improved in several aspects with
  help of Knowledge Plane
• Minimize degradation on links and interference
  from neighbouring APs with automated power
  control and rate adaptation algorithms
• Load management and effective coverage over
  several APs
• Rate adaptation mechanisms
• Traffic scheduling mechnisms to optimize battery
  power
• Trace-driven simulations and small testbed used
  as analysis basis

12
Conclusions

• Context-based management provides means for
  improving the currently complex network
  configuration and control
• Knowledge Plane introduces a new cognitive
  information layer aside the control and data
  planes for intelligent network management
• The principle of Knowledge Plane can be adapted
  and used in several areas and environments aside
  Internet to ensure a common goal, e.g. end-2-end
  QoS
• Together with intelligent and elastic user
  applications, a self-managed and self-organized
  pervasive system can be established

13
References

• A Knowledge Plane for the Internet, David D.
  Clark, Craig Partridge, J. Christopher Ramming
  and John T. Wroclawski, SIGCOMM, 2003
• Sophia An Information Plane for Networked
  Systems, Mike Wawrzoniak, Larry Peterson and
  Timothy Roscoe, ACM SIGCOMM Computer
  Communications Review, Vol 34, Nr 1, Jan 2004
• A Knowledge Plane as a Pricing Mechanism for
  Aggregate, User-Centric Utility Maximization,
  Vladimir Marbukh
• Semantic-Enhanced Distribution Adaptation
  Networks, Bo Shen, Zhichen Xu, Susie Wee and John
  Apostolopoulos, IEEE International Conference on
  Multimedia and Expo (ICME), 2004
• Adding new Components to the Knowledge Plane in
  GMPLS over WDM Networks, Anna Urra, Eusebi Calle,
  J.L. Marzo, IEEE, 2004
• Self-Management in Chaotic Wireless Deployments,
  Aditya Akella, Glenn Judd, Srinivasan Seshan and
  Peter Steenkiste, MobiCom 2005
• Towards a Reliable, Wide-Area Infrastructure for
  Context-Based Self-Management of Communications,
  Graeme Stevenson, Paddy Nixon and Simon Dobson,
  UCD Systems Research Group, Dublin, 2005