Planesforthe Internet Seminar inDistributedComputing ETHZrich - - PowerPoint PPT Presentation

Planesforthe Internet

Seminar inDistributedComputing ETHZürich

SamuelPasquier– psamuel@student.ethz.ch

AboutaVision…

Motivation

• Increasingreliance onIPnetworks
• Increasingcomplexity of IP

networks →Increasingneeds forbetter techniquestomanagethem

Overview

• Motivation
• Existing Planes
• ACleanSlate Design
• Sophia:AnInformationPlane
• The Knowledge Plane
• Conclusion

Today’sExisting Planes

• DataPlane

handles individual packets

• ControlPlane

implements the distributed routing algorithm

• Management(Mgmt)Plane

monitorsthe network configuresDataPlanemechanism and Control

Planeprotocols

• Today’sIPPlanesfarmorecomplex

Forexample:Data Plane’s implementation

next-hop forwarding next-hop forwarding, tunneling, access control, address translation, queuing

Existing Planes

Difficulties

Existing Planes

Difficulties

• Statesconfigured bymultipleentities
• Dependencies between the statesand the logic

updating the states (not maintained automatically)

• Achangetoany partof the configurationcan

easily breakother parts → fragile and complex network

Difficulties

Example – Reachability Control

FrontOffice DataCenter

i1.1 i3.1 i2.1 i2.2 i4.1 i4.2 i3.2 i1.2

?

R1 R3 R2 R5 R4

AFSubnet BFSubnet

Existing Planes

Real Example

• Just amade=up case?
• 24hour Blackout of MajorMicrosoftWeb

sites(January 2001)

Reason:improper configurationof routerbya technician

• Largebreakdown of Swisscom Backbone

network(3weeks ago)

• fficialreason:DDoS Attack

hearsay:improper configuration

Existing Planes

Real Example

"Onemisconfigured routerorunforeseen eventcantakedownanetwork.Ithastodo witharchitecture,buttherearealotof unknownthingsthathappen."

KeithLowry, vicepresidentofsecurityoperations fornetworkconsultingfirmPilotNetworkServices

ACleanSlate Design

ACleanSlate Design

DesignPrinciples

• Network=level objectives

Today:objectivesexpressed inlow=level

configuration → error=prone!!!

Principle:Objectivesshould be expressed

separately from networkelements

• Network=level objectives

Objectivesconcerning performance,reliability

and policy forexample:Reachability Control "Donot allow hostsinsubnet BFtoaccess the accounting servers insubnet AF"

R1 R3 R2 R5 R4 AFSubnet BFSubnet

ACleanSlate Design

DesignPrinciples

ACleanSlate Design

DesignPrinciples

ACleanSlate Design

DesignPrinciples

• Network=level objectives

Purpose:abstraction→ arobust network

ACleanSlate Design

DesignPrinciples

• Network=wide view

Today:measurement supportadded asan

afterthought

Principle:coherent snapshot of the stateof each

networkcomponent

Timeand space dependant Informationabouttopology,traffic,events Purpose:arobust network

ACleanSlate Design

DesignPrinciples

• Directcontrol

Today:Mgmt Planehas only indirectcontrol

• ver the network

Principle:only Controland Mgmt system

responsible forsetting all states of the Data Plane

Nodecision logic should be hardwired Purpose:meet Network=level objectives

ACleanSlate Design

Proposal

• Propose4Planestoachieve these objectives
• Extreme designpoint:Mgmt and Controldecisions

madeinalogically centralized fashion

ACleanSlate Design

Proposal

• Decision Plane

makes all the decisions (reachability,load

balancing,…)

needs Network=wide view usesstandardalgorithms toturn Network=

level objectives into packet=handling state

• DisseminationPlane

Communicationsubstrate Disseminationpaths independant from data

path

ACleanSlate Design

Proposal

Decision Plane Data Plane

Managementinformation states

Discovery Plane

ACleanSlate Design

Proposal

• Discovery Plane

Discovery of physical componentsand their

relationships

creates identifiers torepresent them

• DataPlane

handles individual packets

(based onstategiven from Decision Plane)

ACleanSlate Design

DesignPrinciples reached?

• Network=level objectives

Decision Planesatisfies network=level objectives

• Network=wide view

Decision Planeoperates onnetwork=wide view

• Directcontrol

Decision Planehas directcontrolover the

• peration of the DataPlane

ACleanSlate Design

Evaluation

± Complexity ± Robustness ± Security Solving orcreating problems? Ormoving problems?

Sophia AnInformationPlane

Sophia

AnInformationPlane

• Adistributed system
• incorporates 3functions

Collecting informationaboutnetworkelements Evaluating statements aboutnetwork’s state Reacting according toconclusiondrawn

←

Sophia

BuildingaNetwork=wide view

• Goal:make statements aboutthe overall

networkstateand behavior

• usesProlog

+ Noaprioriassumption aboutpossiblestatesof the system + Programms and dataareequivalent

• Timeand locationexplicitly partof every term

Sophia

Network=wide view = Performance

• Caching introduced

Tradeoff between most up

= to = d ate valueand computationlatency

Ability tospecify evaluation times inthe past

• Scheduling

Ability tospecify evaluation time inthe future

Sophia

Network=wide view = Performance

• Evaluationplanning

Analogy:Query planningindatabase systems Maybe better toevaluate some expressionsat some

specific locationsand times becauseof dependencies

Canrewriteoriginalexpressiontoobtain many

componentswhich can benefit from separate planning

Sophia

Evaluation

+ Distributed + Explicitintroductionof time and space + Appropriate useof declarative programming language – Incomplete model →

• Claimed tobe anincarnationof the

Knowledge PlaneforPlanetLab…

The Knowledge Plane

The Knowledge Plane

Overview

Pictureuntil now

The Knowledge Plane

Overview

Picturewith the Knowledge Plane ?

The Knowledge Plane

PossibleAttributes

• Incommon with Sophia and the CleanSlate Design:

Network=wide view Unifiedapproach Sensor=Actuator principle

• New:

Explicitencoutering forcompositional

consideration

example:merging perspectivesand activities of two previous unconnected networks

The Knowledge Plane

PossibleAttributes

• New:

Knowledge produced,managed and consumed at orbeyond "traditional" edges of the network

Sufficient?

Source:Niko Matsakis

The Knowledge Plane

PossibleAttributes

= foundation of the Knowledge Plane = operates effectively inthe faceof generality (infaceof newtechnologies,newapplications) = makes judgement inpresence of partialor conflicting informations

The Knowledge Plane

PossibleAttributes

= recognizes and mediates conflicts inpolicies and goals = performs optimizations,too complicated for humans,inhigh=dimensional environments

The Knowledge Plane

Why aCognitiveSystem?

• Traditionalalgorithmic approaches toadaptivity

areunlikely toprovide the required level of sophistication

Becausethey require:

= complete information(ingeneral) = precise problem formulations = relatively static environment

The Knowledge Plane

Characteristics

• The CognitiveSystems (CS)should be ableto…

… learn and reason:

= CSimproves when newsituationsarerecognized,new actionsperformed: the knowledge basegrows inuseful ways

… gainexperience and trust:

= CSis ultimately abletorecognize problems and toact

• nits own

The Knowledge Plane

AnArchitecture?

• Speculative ;=)
• Relevantpoints:

Distributed Constraint driven (system may adopt any behavior

which is not specifically constrained)

Bottom=up (composition/decomposition of simple

entities to/from complex entities)

The Knowledge Plane

Evaluation

?

• Away tohandle the complexity of networks(?)
• Representing knowledge is already difficult
• Reasoning aboutit even moredifficult
• Would need cognitiveperformance
• utperforming human capabilities

Looking at actual stateof AI…

The Knowledge Plane

Evaluation

?

The Knowledge Plane

Evaluation

• ?
• Implies amodel of trust and persistentrobust

expressionof identity

• Dowe know who we are?
• Will one day acomputerknow who it is?

Who totrust? Lookingattoday’spolitics and society complexity…

The Knowledge Plane

Evaluation

• Work at avery high=level of abstraction

(the right one?)

• Courageof addressing unavoiding issues

Who would controlthis unified system? Aretraditional algorithms abletohandle such

complexity?

Conclusion

• Visionsaboutnewsolutionsfornetwork’s

control&management

• Unified approaches proposed
• Criticism against actual statejustified (?)
• Solutionsnot (fully)successfuls
• Anintelligentnetworkorastupid network

managed byintelligenthuman beings?

Thanks foryour attention

References

• "ACleanSlate4DApproachtoNetworkControl

andManagement“,A.Greenberg,G.Hjalmtysson, D.A.Maltz,A.Myers,J.Rexford,G.Xie,H.Yan, J.Zhan,and H.Zhang

• "Sophia:AnInformationPlaneforNetworked

Systems",M.Wawrzoniak,L.Peterson,and T. Roscoe

References

• "AKnowledge Planeforthe Internet",D.D.Clark,

C.Partridge,J.C.RammingandJ.T.Wroclawski

• http://www.news.com/Internet-companies-

begging-for-attack,-experts-say/2009-1001_3- 251622.html

• http://www.heise.de/newsticker/suche/ergebnis?rm

=result;q=fehlkonfiguration;url=/newsticker/meldu ng/99399/;words=Fehlkonfiguration

Reactions – Questions?

• AboutScienceorSciencefiction?
• Which idea(s)would you keep?
• Which idea(s)would you throw away?
• Avisionoranightmare?