INTRODUCTION TO AUTONOMIC PHARMACOLOGY: Part V Actions of autonomic nerves:
The State of the Art in Network Management for the Future Internet John Strassner Chairman,...
-
Upload
erin-sharp -
Category
Documents
-
view
215 -
download
0
Transcript of The State of the Art in Network Management for the Future Internet John Strassner Chairman,...
The State of the Art in Network Management for the Future Internet
John StrassnerChairman, Autonomic Communications Forum
Director, Autonomic Research, TSSG
Professor, POSTECH
ManFI KeynoteJohn Strassner
Page 2
Agenda
Introduction and MotivationSummary of Management Problems in the Current Internet Incremental, Evolutionary, and Revolutionary ApproachesState of the Art ReviewThe Real Problems of ManagementA Way Forward
ManFI KeynoteJohn Strassner
Page 3
The Digitization of Everything
ManFI KeynoteJohn Strassner
Page 4
The Explosion of Broadband
ManFI KeynoteJohn Strassner
Page 5
Smarter Devices
ManFI KeynoteJohn Strassner
Page 6
NGN Architecture is About Convergence
“Convergence: The Act of Converging and Especially moving toward union or uniformity…” Merriam-Webster Online Dictionary
Seamless Communications requires convergence of multiple technical areas. Seamless Mobility networks are concerned with:
Service Convergence
Device Convergence
Network Convergence
Service Convergence
Device Convergence
NetworkConvergence
ManFI KeynoteJohn Strassner
Page 7
NGN Applications Need Convergence Too!
Seamless Experience requires convergence of multiple business areas. Seamless Mobility applications are concerned with:
Billing Convergence
Security Convergence
Interface Convergence
Management Convergence
Context Convergence
Security Convergence
Interface Convergence
ContextConvergence
BillingConvergence
Management Convergence
ManFI KeynoteJohn Strassner
Page 8
Seamless Mobility Vision…
Seamless Mobility 101 Set of solutions to give the user the experience of being connected
anywhere, anytime, to anything, with any service
“Seamless” emphasizes continuity of experience across multiple spatial
domains, devices, network protocols and access modes
“Mobility” is the next phase of the internet revolution that allows users to
communicate and manipulate information regardless of location
Easy, un-interrupted access to information, entertainment, communication, monitoring and control
Seamless mobility is a framework architecture that enables devices and networks to interoperate using compatible, reusable software
ManFI KeynoteJohn Strassner
Page 9
Seamless Mobility Architecture
Service Delivery Platform
Application &Content Servers
Cellular 3GMobile Broadband Wireless Access
Cable/DSL/Fiber
Core IP Network
Communication Gateway
Communication Gateway Communication
Gateway
Softswitch
Wireless Services Manager
ManFI KeynoteJohn Strassner
Page 10
Agenda
Introduction and MotivationSummary of Management Problems in the Current Internet Incremental, Evolutionary, and Revolutionary ApproachesState of the Art ReviewThe Real Problems of ManagementA Way Forward
ManFI KeynoteJohn Strassner
Page 11
Internet Architectural Limitations (1)
Two fundamental problems exist in its design• Architectural limitations • Inability to relate business needs to network services offered
For this talk, we are interested in an even more fundamental limitation: lack of manageability
ManFI KeynoteJohn Strassner
Page 12
Internet Architectural Limitations (2) Is Too Simple Just Too Stupid?
• The core does not understand the content or purpose of the data that it is carrying, or what each traffic pattern signifies» It can easily accommodate the deployment of new applications» It cannot easily detect when something has gone wrong
• Lack of Service Interoperability»Approaches such as DiffServ define, but do not enforce, the definition of
traffic classification and conditioning, impeding interoperability» Interaction problems between the different network layers are still
unresolved, and the management of such services is still open
• Scalability»The dramatic increase in different types of local networks all point to the
need to support a variety of communication methods that allow interconnecting many different types of devices
ManFI KeynoteJohn Strassner
Page 13
Internet Architectural Limitations (3) Is Too Simple Stupid? Part 2…
• “Smart Edge, Dumb Core”»Advantage: the Internet can easily deploy new applications»Disadvantage: it cannot easily detect when something has gone wrong
• The Internet can run over essentially any network technology (due to its simple use of that technology)»No way to utilize special features for a particular application»Very difficult to take advantage of new business opportunities enabled by
various disruptive technologies and applications.
• The current Internet lacks mobility support»Current Internet naming system is based on the host address, which is
inefficient for mobile and multi-homed devices»Current solutions, such as HIP, do not take into account context-aware
applications, where message communication depends on context.
ManFI KeynoteJohn Strassner
Page 14
Internet Architectural Limitations (4)
Architectural limitations include Trust (or lack thereof)
• Trusting => prone to attack and manipulation by malicious users
• E2E design principle assumes that end-points can communicate freely, making anonymity and privacy very difficult to achieve
• Security is currently isolated in selected protocols, not the system
• Without monitoring capability, external applications must be used to diagnose problems as well as to identify new attacks
• Exacerbated by the use of multiple administrative domains, each enforcing its own set of regulations and constraints
• Existing patches, such as firewalls, violate the original goals of the Internet design, and move it from an open network to a closed or partially closed network.
ManFI KeynoteJohn Strassner
Page 15
Internet Architectural Limitations (5)
Inability to relate business needs to network services
• Example: given an SNMP alarm, how can one define the set of SLAs for each customer that is adversely affected?
• Current network management data does not contain business or system information»This means that network management applications must instead infer
system and service problems
»How can this overcome the heterogeneous programming languages and models with different semantics used?
»Given the above, how can one integrate disparate data to determine the current state and context?
• It is becoming increasingly difficult for system integrators to gather the information that they need
ManFI KeynoteJohn Strassner
Page 16
Internet Management Deficiencies
Must separate data, control, and management layers• Any additional layers must be separated as well
Abstraction• Network device configuration data is complex!• Need to reduce the complexity seen by the human (or machine)
manager without limiting control
Cross-domain management• Administrative policy differences• Social difficulties and privacy concerns• Cooperation on quid pro quo basis
ManFI KeynoteJohn Strassner
Page 17
Agenda
Introduction and MotivationSummary of Management Problems in the Current Internet Incremental, Evolutionary, and Revolutionary ApproachesState of the Art ReviewThe Real Problems of ManagementA Way Forward
ManFI KeynoteJohn Strassner
Page 18
Three Approaches Incremental approach
• Current status quo, characterized by the plethora of point solutions that violate the architectural principles of the current Internet
• Many different problems, especially conflict between point solutions Revolutionary (“clean slate”) approach
• Eliminates existing commitments, restraints, and assumptions, and starts with a new set of ideas
• Advocates a radical redesign of the current Internet architecture Evolutionary approach – a compromise
• Enables new ideas to evolve while simultaneously emphasizing backwards compatibility with the existing Internet
• Current and future networks and networked applications have vastly different requirements; this implies that a single architecture cannot simultaneously meet these different needs
ManFI KeynoteJohn Strassner
Page 19
Definitions
Incremental• System is moved from one state to another with small,
incremental patches; new ideas that are not compatible with the original design are not allowed
Evolutionary• Same as above, except that new ideas that are not
compatible with the original design are allowed.
Revolutionary• System is redesigned from scratch to offer improved
abstractions and/or performance, while providing similar functionality based on new core principles.
ManFI KeynoteJohn Strassner
Page 20
Agenda
Introduction and MotivationSummary of Management Problems in the Current Internet Incremental, Evolutionary, and Revolutionary ApproachesState of the Art ReviewThe Real Problems of ManagementA Way Forward
ManFI KeynoteJohn Strassner
Page 21
Global Environment for Network Innovations (GENI)An experimental infrastructure to validate and demonstrate
networking research launched in August 2005GENI addresses the deficiencies in existing tools and
small-scale experimentation• In principle, this will enable the viability of Future Internet designs to
be validated, especially under realistic future conditions• Concentrates on overarching clean-slate architectural proposals as
well as advances in the basic building blocks of the Future Internet• The GENI research facility is a global experimental facility that
fosters the exploration and evaluation of new networking architectures (at scale) under realistic conditions
ManFI KeynoteJohn Strassner
Page 22
Future Internet Design (FIND)
FIND is a major long-term initiative of the NSF NeTS research program started in 2006
FIND is aimed at designing all aspects of the Future Internet in a timeframe roughly 15 years in the future
FIND seeks to design and build an end-to-end network architecture incorporating emerging technologies using a clean-slate approach
ManFI KeynoteJohn Strassner
Page 23
Situated Autonomic Computing Situated Communications
• Reacting locally to context changes• Considering strategic needs (social or economic, not only technological)
Autonomic Communications• network elements autonomously interrelated and controlled, learning desired
behavior Purpose and Goals
• Identify new “situated” and “autonomic” communication/networking paradigms in a longish (10-20 years) time frame
• Combine technological and socio-economic research and instil into networks• Communication/networking should become task- and knowledge-driven• To define a self-organising communication network that can be situated in
multiple and dynamic contexts• To define hardware and software that can evolve and create maximal
synergy with other non-technological entities that constitute their context
ManFI KeynoteJohn Strassner
Page 24
Situated Autonomic Computing (FP6) ANA
• Novel network architecture (beyond IP) enabling flexible and autonomic formation of network nodes according to working, economic and social needs; focused on adaptation and reorganization of the network
BIONETS• Biologically-inspired approach (from nature and society) to localized
autonomic communication services without central control, allowing high-level services to evolve spontaneously
CASCADAS• Defining a new generation of highly distributed, pervasive, situation-aware,
semantically self-organising communication-intensive services• Focus on self-similarity, autonomic component-ware
HAGGLE• Cross-layer network architecture exploiting intermittent connectivity• Supporting opportunistic networking paradigm (delivery of messages based
on store and forward exploiting situated information)
ManFI KeynoteJohn Strassner
Page 25
4WARD Overview
Research on technical aspects of the Future Internet and its relations to non-technical aspects like regulation and business modeling
Common elements:• Network virtualization, service provisioning resources, new forms of
connectivity
Novel elements:• Network of information - focusing on information objects• Information-centric, not device-centric, communication• Seeks to integrate non-technical and technical work areas: ‘business
evolution’, ‘socio-economic environment’ and ‘regulation’.• Autonomic self-organizing management plane
ManFI KeynoteJohn Strassner
Page 26
4WARD Network of Information
Information object model• including object meta-data, multiple representations at the bit-level,
versioning, support for live streaming and support for physical objects Information object integrity
• Object identification and verification of authenticity independent of its location Object search
• Google for objects Object lookup/resolution
• Mechanisms for finding the “best” copy of the object, given a name or identifier for the object, private scope as well as public/global scope
Object distribution• Mechanisms for information object routing and caching, including replication
and synchronisation, and optimisation of delivery Object storage
• Management of storage for objects
ManFI KeynoteJohn Strassner
Page 27
Autonomic Internet Overview (FP7) Creates a management resource overlay with autonomic
characteristics for the purposes of easy, fast and guaranteed service delivery.• Develop open software infrastructure and tools to support composition and
execution of fast and guaranteed services• Based on using virtual network resources and Policy-Based Management to
describe and control services• Uses Ontologies, Information, and Data models to facilitate the Internet
service deployment in terms of programmable networks facilities supporting the Future Internet
• Develop a set of service-centric network APIs to abstract the heterogeneity of multiple types of access and core networks
Self-Management through• Using and extending FOCALE architecture• Building analogous knowledge plane and control loops• New network APIs designed to take advantage of model and virtual
device/service abstractions
ManFI KeynoteJohn Strassner
Page 28
Mapping
…
Mapping
…
OrchestrationPlane
VirtualisationPlane
Semantic Bus
Federation Services
…User Interface Services
Business Goals
Customer Needs
Models andOntologies
DistributedOrchestrationComponent
Autonomic Management System
Management Plane
Knowledge Plane
Developer Interface
Operator InterfaceService
LifecycleManagement
Lifecycle Management Services
Intra-System View
Oth
ers
Dis
trib
uti
on
Dis
trib
uti
on
Fed
erat
ion
Fed
erat
ion
Neg
oti
atio
nN
ego
tiat
ion
System View
Oth
ers
Dis
trib
uti
on
Dis
trib
uti
on
Fed
erat
ion
Fed
erat
ion
Neg
oti
atio
nN
ego
tiat
ion
Oth
ers
Oth
ers
Dis
trib
uti
on
Dis
trib
uti
on
Dis
trib
uti
on
Dis
trib
uti
on
Fed
erat
ion
Fed
erat
ion
Fed
erat
ion
Fed
erat
ion
Neg
oti
atio
nN
ego
tiat
ion
Neg
oti
atio
nN
ego
tiat
ion
System View
Oth
ers
Dis
trib
uti
on
Dis
trib
uti
on
Fed
erat
ion
Fed
erat
ion
Neg
oti
atio
nN
ego
tiat
ion
Oth
ers
Oth
ers
Dis
trib
uti
on
Dis
trib
uti
on
Dis
trib
uti
on
Dis
trib
uti
on
Fed
erat
ion
Fed
erat
ion
Fed
erat
ion
Fed
erat
ion
Neg
oti
atio
nN
ego
tiat
ion
Neg
oti
atio
nN
ego
tiat
ion
vCPI vCPI vCPI
vSPI
vCPI vCPI
Autonomic Management System
Management Plane
Knowledge Plane
Physical Resource
Component
Inter-System ViewAll Orchestration Modules
End-User Virtualised Services
End-User Composite Virtualised Services
ManFI KeynoteJohn Strassner
Page 29
Agenda
Introduction and MotivationSummary of Management Problems in the Current Internet Incremental, Evolutionary, and Revolutionary ApproachesState of the Art ReviewThe Real Problems of ManagementA Way Forward
ManFI KeynoteJohn Strassner
Page 30
Current Management Limitations
Network infrastructure is developed and deployed first• Management and security are added separately later• Thus, functions which need management cannot be designed or
tested until management is added• No opportunity to simulate or test before deployment
Lack of vendor- and technology-neutral models and APIs mean that even a local management approach cannot be designed beforehand
Inability to incorporate new or changed information Inability to customize services offered according to
business and environmental conditions/constraints
ManFI KeynoteJohn Strassner
Page 31
Key Management RequirementsSecurity and trustworthiness in a distributed environment
• by embedding security and trust rules in network
Management functionality at modeling and design phases• Must support different levels of management compliance
»Default/essential vs. nice-to-have/extended vs. /optional• Monitoring and accounting functions must be embedded capabilities
through a uniformly accessible • Ability to support evolutionary management needs in response to
new device functionality, customer service requirements, etc.• Separation of management, control, and data traffic
Pre-defined management functions help business• Reduced integration costs• Shorter development and testing life cycles
ManFI KeynoteJohn Strassner
Page 32
Additional Management RequirementsSupport for four key types of functionality
• Operations – Keep the network running smoothly, monitor for alarms• Administration – Keep track of network assets, who uses what, etc.• Maintenance – Repairs, upgrades, planning• Provisioning – Network device and service configuration
Support for additional functionality• Help identify, diagnose, fix problems (reactive management)
»Event and alarm correlation/filtering (traditional view)»Use of abstraction to reason about root cause (emerging view)
• Avoid problems in the future (proactive management)»Observe and extrapolate performance trends»Support problem prediction
• Move beyond traditional manager-agent architectures to support flexibility better suited to application needs
ManFI KeynoteJohn Strassner
Page 33
Nice to Have Management RequirementsManagement interfaces must be flexible
• Dedicated protocol has advantages and disadvantages• Freedom to use APIs, logic, or other approaches that are better
suited to the needs of the application
Management interfaces should be consistent• Do we really need different protocols for configuration vs. accounting
vs. security vs. …?• Is there really a need for different management interfaces that
express different views of the same data?Management data should be consistent
• Exchange and reuse of common data is mandatory
ManFI KeynoteJohn Strassner
Page 34
Agenda
Introduction and MotivationSummary of Management Problems in the Current Internet Incremental, Evolutionary, and Revolutionary ApproachesState of the Art ReviewThe Real Problems of ManagementA Way Forward
ManFI KeynoteJohn Strassner
Page 35
FIXEDKnowledge Base
Manages
Typical Current Management Approach
Management System
Pre-Defined LogicFunctionality
Receives vendor-specific commandsSends vendor-specific data
ManFI KeynoteJohn Strassner
Page 36
ProductionNetwork
DedicatedManagementNetwork
ProductionNetwork
Shared management and data traffic
Management trafficProduction traffic
Legend:
Typical Management NetworksSeparate management and data traffic
ManFI KeynoteJohn Strassner
Page 37
Shared Management Typically UsedAdvantages of Shared Management
• Significant savings in cost and overhead• Often not practical to build separate communication• More communication lines may lead to additional security risks• Often simpler to plan one network
Disadvantages of Shared Management• Reliability decreased since management traffic may be blocked• Performance impacted since management traffic can interfere with
mission-critical data traffic
»Of course, this assumes that management is not mission-critical
• Security impacted, since users and administrators are performing different functions over the same network infrastructure
ManFI KeynoteJohn Strassner
Page 38
Next Generation Management Interfaces Network nodes will vary in functionality, programmability, persistence,
and dependability Services will rapidly increase and diversify, so management data model
must be extensible• Specify management objects from different viewpoints (e.g., business vs.
technical)• Software must be responsible for translating high-level goals into appropriate
strategies for lower level implementation Separation of data, control, and management functions Inclusion of “inferencing” functions to enable self-awareness and
reasoning to be performed• Management depends not on just observed and measured facts, but also
inferred results• Number of data sources and their inherent diversity in defining management
and operational data make their integration a difficult problem to solve
ManFI KeynoteJohn Strassner
Page 39
IBM MAPE Approach
Autonomic Element Autonomic Element Autonomic Element
ManFI KeynoteJohn Strassner
Page 40
CASCADAS ACE
ManFI KeynoteJohn Strassner
Page 41
FOCALE ACE
Autonomic Computing Element
Autonomic ManagerContext-Aware
Policy ServerLearning and
rEasoning
Observe Compare Act
Model-Based Translation
DistributedComponents
DistributedComponents
Foundation:Finite State
Machine
Models and Ontologies
ECB
ManFI KeynoteJohn Strassner
Page 42
FOCALE Control Loops
Observe Compare
Decide
Context
Normalize
Act
Policy Rules Inner Loop
Outer Loop
Adaptation
ManFI KeynoteJohn Strassner
Page 43
Semantics in FOCALE
Raw Eventsand/or Data
Information Objects
Knowledge Concepts
Managed Resource(s)
Information ModelsInformation
ModelsInformation Models
OntologiesOntologies
Ontologies
Information Model Mapping Logic
Ontology Model Mapping Logic
Harmonization Logic
XML
XML
XML
Object Construction and Semantic
Augmentation Logic
Autonomic Processing Engine
Semantic XML Objects
XMLXML
Se
ma
ntic
Bu
s
ManFI KeynoteJohn Strassner
Page 44
The Inference Plane
Data Plane
Control Plane B
Inference Plane
Autonomic ManagerAutonomic Manager
Policy Server
Observe
Learn
Actions
DEN-ngModels and Ontologies
Foundation:Finite StateMachines
Compare
Policy ServerPolicy ServerPolicy Server
ObserveObserveObserve
LearnLearnLearn
ActionsActionsActions
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
Foundation:Finite StateMachines
Foundation:Finite StateMachines
Foundation:Finite StateMachines
CompareCompareCompare
Control Plane A
Network Management Plane Network
Network Network A
ManFI KeynoteJohn Strassner
Page 45
FOCALE Service-Aware Domain
Physical Device (part of Data Plane)
Administrative Domain
FOCALE
Virtual ServiceVirtual ServiceVirtual Service
Virtual Device (part of Data Plane)Virtual Device (part of Data Plane)Virtual Device (part of Data Plane)
Fusion Logic
Mgmt & Ctrl
OrchestrationOrchestration
ManFI KeynoteJohn Strassner
Page 46
ComparisonAspect IBM CASCADAS FOCALE
Static or Adaptive Control Static Adaptive Adaptive Multiple Control Loops Single Single Multiple
Policy Driven Control Loop Yes No Yes Policy Languages One None Multiple
Policy Languages for Different Constituencies
No No Yes
Dynamic Knowledge Base No Yes Yes Data Scalability Limited by
sensors Scalable Scalable
Accommodates Heterogeneous Data
Limited by Common Base
Event
No mechanism to deal with this
Yes through model-based translation
Data Complexity Limited to data that can be
instrumented
Specific plugins can be added for new data
Uses patterns, models, and ontologies to
parse data Data Semantics Encoded No Yes Yes
ACE Complexity High Low Low ACE Components
Distributed No Yes Yes
Uses Ontologies No Could in Future Yes Semantic Matching No Yes Yes
Self-Organizing No Yes Yes Uses Self-Model No Yes Yes
Publishes Self-Model? No Yes No State-Driven No Yes Yes
Adaptable State Machine No Yes Yes Model-Based Supervision No Yes Yes
Autonomic Behavior Loosely or Tightly Coupled
Tightly Coupled
Both Both
Context-Aware No Yes Yes Supports Emergent
Functionality No Yes Yes
Communication Mechanism Pre-Defined Interfaces
Messages and pre-defined interfaces
ECB messaging and pre-defined
interfaces
ManFI KeynoteJohn Strassner
Page 47
PolicyService
Event Service
SecurityService
Discovery Service
Context Service
AnalysisService
…
Autonomic Management Environment
PolicyServicePolicy
Service
Event Service
SecurityServiceSecurityService
Discovery Service
Discovery Service
Context ServiceContext Service
AnalysisServiceAnalysisService
…
Autonomic Management Domain
Managed ResourceManaged Resource
Model-Based Translation LayerModel-Based Translation Layer
Autonomic Manager
Autonomic Computing Element
Context-AwarePolicy Server
Context-AwarePolicy Server
Context-AwarePolicy Server
ObserveObserveObserve
Learn andrEason
Learn andrEason
Learn andrEason
ActionsActionsActions
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
Foundation:Finite StateMachines
Foundation:Finite StateMachines
Foundation:Finite StateMachines
CompareCompareCompare
Managed ResourceManaged Resource
Model-Based Translation LayerModel-Based Translation Layer
Autonomic Manager
Autonomic Computing Element
Context-AwarePolicy Server
Context-AwarePolicy Server
Context-AwarePolicy Server
ObserveObserveObserve
Learn andrEason
Learn andrEason
Learn andrEason
ActionsActionsActions
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
Foundation:Finite StateMachines
Foundation:Finite StateMachines
Foundation:Finite StateMachines
CompareCompareCompare
PolicyServicePolicy
Service
Event Service
SecurityServiceSecurityService
Discovery Service
Discovery Service
Context ServiceContext Service
AnalysisServiceAnalysisService
…
Autonomic Management Domain
Managed ResourceManaged Resource
Model-Based Translation LayerModel-Based Translation Layer
Autonomic Manager
Autonomic Computing Element
Context-AwarePolicy Server
Context-AwarePolicy Server
Context-AwarePolicy Server
ObserveObserveObserve
Learn andrEason
Learn andrEason
Learn andrEason
ActionsActionsActions
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
Foundation:Finite StateMachines
Foundation:Finite StateMachines
Foundation:Finite StateMachines
CompareCompareCompare
Managed ResourceManaged Resource
Model-Based Translation LayerModel-Based Translation Layer
Autonomic Manager
Autonomic Computing Element
Context-AwarePolicy Server
Context-AwarePolicy Server
Context-AwarePolicy Server
ObserveObserveObserve
Learn andrEason
Learn andrEason
Learn andrEason
ActionsActionsActions
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
DEN-ngModels and Ontologies
Foundation:Finite StateMachines
Foundation:Finite StateMachines
Foundation:Finite StateMachines
CompareCompareCompare
ManFI KeynoteJohn Strassner
Page 48
Future Research
We are designing a new network architecture from an information-centric approach• Rather than from a device-centric approach
The information is the focus• Not where we get it
Brings several benefits• Better large-scale distribution efficiency• Better performance an reliability (also for non-dissemination
applications)• Supports dynamic adaptation and orchestration• Supports ability to reason about data
Ongoing work on information modelling, object naming and the architectural framework
Questions?
Questions?
“Create like a god. Command like a king. Work like a slave”- Constantin Brancusi