Autonomic Computer Systems and their potential application to Road Network Management

1
Autonomic Computer Systems and their potential
application to Road Network Management Lee
McCluskey Dept of Informatics
2
Future / current problem with complexity of
systems
jpeg from an address to The 3rd IEEE
International Conference on Autonomic Computing
by David Bartlett
3
Future / current problem with complexity of
systems

• As (computer) systems are getting more complex,
  they are getting much harder to manage, in
  particular to ..
• Maintain
• Fix
• Configure / Re-configure
• Secure
• Initial requirements drift, environments change.

4
Autonomic Nervous Systems

• Human breathing, heartbeat, temperature, immune
  system, repair mechanisms are all to a great
  extent controlled by our body without our
  conscious management
• When we are anxious, frightened, ill, injured etc
  all these bodily functions have evolved to react
  appropriately.

5
Autonomic Computing

• Apparently the term was coined by IBM in 2001,
  now there is an international conference (ICAC
  2004,5,6,7)
• AC/ICAC have attracted a very wide range of
  computing research and approaches, largely from
  the artificial intelligence community
• Aim to make complex systems self-managing, ie
• Self healing
• Self maintaining
• Self configuring
• and thus deal with the problems outlined above.
• Up until recently, research into individual
  autonomic properties was tackled piecemeal. Some
  systems (eg computer operating systems ..) have
  some elements of these features already

6
Capabilities required for AC

• The computer system may needs to be able to
• Self-Evaluate detect and identify problems
  (security breaches, SLA shortfalls, bugs)
• Learn tune, auto-debug, change code, seek new
  knowledge / services when required
• Plan creating strategies/schedules for dealing
  with problems
• These generally require the system to have an
  explicit set of requirements /SLAs (so can
  evaluate) and the semantics of components of the
  system

7
A web approach to AC ..

• Nowadays the field of SOA emphasises the idea of
  putting components together to create systems ..
  EG
• Systems are made up of compositions of web
  services
• Web services are marked up with semantics (eg
  using OWL-S)
• Requirements are like goals that compositions
  of web services must achieve, they are configured
  automatically by automated planning programs
• As new services come on stream, or services are
  unavailable, or are retired, self-maintenance
  means replacing old with new but equivalent
  semantics.
• If more than evolution is required,
  self-configuration is down to re-planning, where
  a new solution is constructed from primitive web
  services.
• (cf PhD student Faisal Mustafa who is looking
  into this)

8
A lot my research has really been AC!

• See my website http//scom.hud.ac.uk/scomtlm/ for
  details
• FAROAS 1992-4
• Create formalisation of functional requirements
  (the CPS), automate the solution / code writing
  code self-maintenance
• IMPRESS 1996-9
• Use training data to test the CPS-generated
  code, learning processes both find the
  requirements that are to blame for errors, and
  invent a requirements fix to satisfy the training
  data. A new program is automatically derived
  auto-bug fixing and self-learning
• PLANFORM 2000-3
• Produce platform to ease the process of automated
  planning self-management

9
AC application Road Network Management

• Complex data gathering, knowledge extraction,
  planning and control application
• Critical to congestion control, incident
  management, road use optimisation

10
Road Network Management
many systems for data gathering (various types of
traffic detectors, cctv) many ways to influence
and control traffic lights, variable speed
limits, variable message signs, local radio,
satnav (near future?) System control involves
forming and executing a strategy to control and
hence optimise the network
11
Road Network Management Current Situation

• Main problems
• Road management experts scarce
• Many different systems, not interoperable
• Systems for different areas (HA,Urban)
• Usual problems of maintenance, configuration
• Road management experts scarce
• However
• Data services becoming standardised
• Road strategies semi-formalised
• Some intelligent systems exist but on a small
  scale

12
A Solution with AC properties
Autonomic Properties
Persistent Knowledge
Domain Model
configure
learn
Declarative Description of Plan
Outlines Policies Requirements Network
Characteristics and Geometries Actions Network
Affecting Actions Info Gathering
maintain
Plan construction, plan tuning
RNMS
Real Time Network Information
manage
heal, repair
Transient Knowledge
plug-in service semantics
Available Data Services
ROAD NETWORK
13
Road Network Management Some of the functions
of our AC approach

• Selfhealing/ repairing (Information fusion)
• Self-management (of the Road Network)
• Self-maintenance (of Domain Model eg when a new
  road is inserted, or a new service becomes
  available)
• Self-configuration (eg new Road Network)

14
Road Network Management Some of the benefits of
our AC approach

• Enforce an application-independent declarative
  semantics for plans, policies, network
  geometries, information services etc. Service
  providers can plug-in to this, and new services
  can be plugged-in..
• A great deal of s/w maintenance will amount to
  knowledge editing
• Planners and solvers will improve automatically
  as the knowledge/services available improves
• Real time information will rely on a minimum of
  network sensors

15
More detailed system architecture
16
Summary

• Computer systems are becoming too complex to
  manage efficiently
• AC brings together R and D into giving (computer)
  systems self management properties - using AI
  techniques such as planning and learning
• We have identified an application which AC would
  benefit
• Self management of the overall system
• Self maintenance, configuration, and repair of
  some of the embedded system components