From Active Networks to Cognitive Networks

1
From Active Networks to Cognitive Networks

• Manolis Sifalakis
• m.sifalakis_at_lancaster.ac.uk
• University of Lancaster

2
Overview

• The Vision Cognitive Networks (a long term aim)
• Motivation Case scenario(s)
• Structure
• Fundamental requirements
• The role of Active Networks
• A Service Deployment Toolkit (a short term
  objective)
• Key principals
• Toolkit functionality
• The benefits
• Example deployment

3
Part A The Vision Cognitive Networks

• A new generation of service networks
• Properties
• Cognitive actions based on Reasoning
• Autonomic operation
• Adaptive functionality
• Self Manageability
• Aims
• Resilient servicing
• Service Dependability (even when hardware fails)

4
Case Scenario Weekend DoS Attack

• Problem Description
• Scientists run lengthy experiments using network
  resources throughout the weekend
• Administrators not working in weekends
• DoS attack launched on Fri night
• No support service disruption for 3 (at least)
  days
• Observation
• The vast majority of attacks today are recipe
  style and follow specific patterns
• Cognitive Solution
• Train a classification system to detect the
  attack patterns
• Find a suitable software to respond to the attack
  detection
• Use active networks to dynamically deploy/upgrade
  the defence system whenever/wherever needed

5
Case Scenario Roaming Multimedia User

• Problem Description
• Mobile user attends a confidential
  videoconference
• At every visited network needs sustainable media
  quality and security services
• Some or all of these services, cannot be
  available in every network and on a per user
  basis
• Cognitive solution
• Provide a set of basic active service components
  for proxy, encryption, anonymizing, MPLS, and
  transcoding
• Develop an intelligent personalised agent
• to sense/investigate the visited environment
  and
• combine and deploy the modules in correct order
  and correct locations
• Use active networks to deploy them dynamically
  wherever/whenever required

6
Case Scenario Sensor Network Deployment

• Problem Description
• Randomly laid semi-mobile devices
• Collection of environmental data in a natural
  catastrophes sensitive environment
• Need for auto-configuration, integration in the
  mesh, reliable, secure safe exchange of data
• Varying environmental conditions impact the
  network performance
• Different network setups perform better in
  different environments gt need for adaptive
  solutions
• Cognitive solution
• Use some AI-based context aware elements to
  detect assess the environment changes select
  suitable protocol suites and network
  configurations
• Design a p2p system for the synchronisation and
  coupling of the network devices at the service
  level
• Use active networks for the on-line and dynamic
  configuration and update of the coupling elements
  and control service modules

7
The Key Technologies

• A combination of
• A.I. strategies
• Policy enforcement systems
• (Mobile) agent technologies
• Active Programmable networks
• P2P systems
• Semantic based services/languages/tools
• Context aware services
• etc

8
How ? A Layered Structure

• Cognition is twofold
• User Servicing
• Network Management

• Active Networks
• Collection of information
• Action implementation

9
Fundamental Requirement

• Key Requirement
• Decoupling of service management from
  infrastructure management
• More persistent service provisioning even when
  the underlying infrastructure fails
• Active Networks seem to be the definite enabling
  technology to satisfy the requirement
• On-line adaptability
• Programmability
• Dynamic service (de/re-)composition

10
Problems in Active Network Research

• Multiple diverse platforms Non interoperable!
• Many specialised architectures Almost none
  generic enough! (maybe FAIN in the future ?)
• Several implementations Most still in the lab!
• Several applications Few that necessitate the
  use of active/programmable technologies
• Lack of frameworks for large scale and
  multi-platform deployment
• Security issues and complexity in management and
  administration
• Funding seems to be gradually finishing!!!

11
Part B A Toolkit for Generic Service
Deployment

• A collection of low-level lightweight tools
  (active services).
• Main Objectives
• Assist the large-scale deployment and
  interoperability of active resources (services,
  platforms, EEs, etc).
• Enable decoupling and abstraction of active
  service deployment from infrastructure management
• Key Functionalities
• Determine the interfacing between active
  resources (platforms, protocols, service
  components)
• Discover and recruit active nodes
• Deploy active service components
• Assist the organisation and management of
  composite services provisioning

12
Active Node Discovery Recruitment

• Organisation of global active resources in
  2-level overlay topology for control path comm.
  (discovery, allocation, coordination)
• Intradomain (intra-AS)
• Interdomain (Inter-AS)
• Full functionality at each level independently
• Designated Active Nodes are the connecting links
  between the 2-levels
• Interdomain level interconnectivity follows the
  AS connectivity pattern
• Distributed network pool model Active nodes that
  cannot serve more requests, leave the overlay

• Issues
• Dynamic, automatic optimal formation of the
  intradomain overlay
• Automatic (s)election of the designated node
• Interdomain connectivity when non active network
  enabled AS is interjected
• Handling of overlay partitioning

13
An Active Proxy well, why not many of them

• An active service per se
• Role
• abstract the service rollout process,
• decouple the service acquisition from the
  service deployment
• Available to users through intradomain
  multicast/anycast
• Benefits
• Single point of trust and control
• Offload the end node from the service rollout
  instrumentation

14
Unified Active Service Deployment

• Need for a generic and abstract service
  deployment interface
• Allocate resources
• (Un-)Install Services
• (Re-)Configure service provisioning
• (De-)Activate servicing
• Must abstract any platform-specific service
  loading mechanism
• ASDP protocol. Ongoing work on a newer version.

15
Organisation Management of Service Composites

• Cooperating service components organise in p2p
  overlays for signalling and control path
  management
• P2P systems use application level (i.e. e2e)
  performance inefficient solutions
• A dynamic kernel level tunnelling mechanism
• Performance efficient, dynamic, low level p2p
  overlay construction
• Application (service) specific routing at kernel
  level
• Unified API shared by all active applications

16
Active Service Mobility Framework

• Mobilisation/migration of running active service
  components between active nodes
• Benefits
• Resists infrastructure failures,
• Enhances user mobility
• Adapts to network weather changes
• Based on the programmable switch approach
• Combines strengths and overcomes limitations of
  active capsule and mobile agent technologies
• Low level forwarding path programmability
• Out-of-band deployment and management
• Also considered in FAIN but not implemented(?)

17
How they all fit together in one EE
18
An Example DoS Detection/Interception Service

• Cognitive logic
• Resource mgmt level (Active Proxies what/where
  to deploy w.r.t. service specification)
• Service level (recognition and classification of
  traffic patterns and counteraction)

• Toolkit functionality
• Distributed, (semi-)mobile agent service
• Migrating sensor elements (Sx) use the service
  mobility framework
• Agent modules (C, Sx, F) form a p2p group
  (control path coordination) using the dynamic
  kernel tunnelling service
• Active resource availability checked thru the
  global network overlay of active resources
• Requests for service deployment, serviced by
  active proxies
• Installation and activation of service components
  uses the unified active service deployment
  interface

19
Related Work

• Mainly from ETH ... But elsewhere as well
• Chameleon
• Netkit
• FAIN
• Alpine
• Etc
• They focus mainly at the interfacing and
  deploying servicing elements (within a platform)
• We try to address problems at a lower level
  Interfacing and cooperation between platforms
• Both complementary as well as necessary

20
Conclusions

• The road to CognitiveNet-Shire passes from the
  ActiveNet-Land ?
• Its a long way and there is still lots of space
  for research
• Large scale deployment of active nodes
  (coordination organisation)
• Platform interoperability
• Security
• Dynamic (on-line) service composition
• etc
• Lots of applications
• User mobility,
• Ad-hoc sensor networks,
• Network/service self-management,
  auto-configuration, survivability
• etc
• Need more imagination, vision funding !!! ?