italy presentation

Marcell

Download

Post to :

URL :

Related Presentations :

Add to Flag Embed

Uploaded from authorPOINTLite

Views: 394

Category: Education

Presentation Description

No description available.

Comments

Presentation Transcript

Some Thoughts on Future Networks:

Some Thoughts on Future Networks Tatsuya Suda Professor, University of California (web) netresearch.ics.uci.edu (email) suda@ics.uci.edu

Outline:

Outline Where are we now? Where are we heading? What do we need to do?

Where are we now?Next Generation Internet (NGI) Initiative :

Where are we now? Next Generation Internet (NGI) Initiative

Slide4:

NSF-9 Major Departments Independent Agencies

NGI　(Next Generation Internet):

NGI　(Next Generation Internet) NSF, DARPA, NASA, NIST, NIH, (DoE) Large Scale Networking working group (in the National Science and Technology Council under the White House) to coordinate multi-agency efforts

NGI　(Next Generation Internet):

NGI　(Next Generation Internet) 3 NGI Goals Goal 1: Basic Network Research Goal 2: NGI Testbed Goal 3: Evolutionary Applications

Slide8:

Pac Bell NAP HAY San Francisco NCAR National Center for Atmospheric Research SDSC San Diego Supercomputer Center HSJ Houston DNJ Denver Ameritech NAP DNG Chicago NCSA National Center for Supercomputing Applications NOR Cleveland CTC Cornell Theory Center PYM Perryman, MD Sprint NAP MFS NAP PSC Pittsburgh Supercomputer Center C A C A C C RTO Los Angeles C A C C AST Atlanta C C C A C C C C C WOR New York City C vBNS Backbone Network Map

California Research Network-2 (CalREN-2):

California Research Network-2 (CalREN-2)

CalREN-2 - Southern Region:

CalREN-2 - Southern Region

U.C. Irvine ATM Network:

U.C. Irvine ATM Network Sparc5 TAXI Sparc5 OC-3 Pentium OC-3 Gigaport OC-3 Pentium OC-3 OC-3 OC-12 OC-3 OC-3 7507 ISI SCCSD ICS Department

STAR TAPScience Technology and Research Transit Access Point:

STAR TAP Science Technology and Research Transit Access Point http://www.startap.net

Canada’s CA*net II Network:

Canada’s CA*net II Network Source: http://www.canarie.ca/frames/n3.html

Europe: TEN-34Trans-European Network – 34 Mbps:

Europe: TEN-34 Trans-European Network – 34 Mbps Source: http://www.dante.net/ten-34/ten34net.gif

APAN - Asia Pacific Advanced Network :

APAN - Asia Pacific Advanced Network

Applications:

Applications Digital libraries, remote operation of medicine, environment, crisis management, manufacturing, basic science, Federal information services

Virtual Temporal Bone:

Virtual Temporal Bone VR model viewers explore using a wand and a special pair of eyeglasses while facing ImmersaDesk surgeons familiarize themselves with the complex structures composing the ear Source: http://www2.sbhis.uic.edu/VRML/vtb0.htm

CAVE Research Network: CAVERN:

CAVE Research Network: CAVERN next generation collaborative networking infrastructure alliance of industrial and research institutions equipped with CAVEs, ImmersaDesks, and high-performance computing resources interconnected by high-speed networks Source: http://www.evl.uic.edu/spiff/covr/vrserver.html CAVE with NICE for Children

SuperComputing’97 Demo from MPI-Garching Gravity Waves from Two Black Holes Colliding:

SuperComputing’97 Demo from MPI-Garching Gravity Waves from Two Black Holes Colliding 3-D Source: http://jean-luc.ncsa.uiuc.edu/Movies/Images

Slide20:

Similar IT project started in Japan last year

Where are we now?Example Internet Traffic Control Schemes :

Where are we now? Example Internet Traffic Control Schemes

Slide24:

Need new scalable techniques/mechanisms traffic control QoS support

Slide25:

Researchers This conference CalREN 2 Experiments QoS Routing Muticast IPv6 IETF QoS Policy Server Architectures MPLS, Traffic Engineering Many other topics being discussed

Current Internet:

Current Internet Philosophy Simple, robust and scalable Complexity pushed to the edges of the network End-to-end flow and congestion control Core-stateless Network Core routers are NOT required to Keep per-flow state accounting or queueing Perform per-flow processing or scheduling Implement per-flow congestion/flow control

Slide27:

Simple and scalable Limited controllability Some shortcomings Congestion collapse Unfair bandwidth allocation

Core-stateful Networks:

Core-stateful Networks ATM and Telephone networks Virtual circuit (VC) based Fixed path Routers or switches keep per VC state (VC forwarding table) and may perform per VC tasks Can guarantee quality of service Complex and expensive to deploy

Core-stateless Networks:

Core-stateless Networks Core routers are NOT required to Keep per-flow state accounting or queueing Perform per-flow processing or scheduling Implement per-flow congestion/flow control Simple and scalable Limited controllability

Example Core Stateless MechanismRandom Early Drop (RED):

P(drop) = 1 - e-bb Example Core Stateless Mechanism Random Early Drop (RED) Drop a new packet with a probability dependent on the buffer occupancy not a fixed threshold Example

Weighted Fair Queueing (WFQ):

Weighted Fair Queueing (WFQ) Flow classifier places packets in per-flow queues (core-stateful approach) Scheduler (e.g. deficit round robin) fairly serves the per-flow queues Packets from flows transmitting at lower rates experience lower delays!

Example Core Stateless Mechanism Core-stateless Fair Queueing (CSFQ):

Example Core Stateless Mechanism Core-stateless Fair Queueing (CSFQ) Packets are labeled by ingress routers with the flow input rate Packets are placed in a single FIFO queue If the queue occupancy is higher than a threshold, Drop-on-input module probabilistically drops packets from flows transmitting at a rate higher than their estimated fair share If the queue occupancy is lower than a threshold, CSFQ is ineffective Packets from all flows experience the same delay!

Current Internet:

Current Internet Philosophy Simple, robust and scalable Complexity pushed to the edges of the network End-to-end flow and congestion control Some shortcomings Congestion collapse Unfair bandwidth allocation

Congestion Collapse:

FTP Source Video Source Dest Dest Bottleneck link Undelivered packets Congestion Collapse

Unfairness due toUnresponsive Flows:

FTP Source Video Source Dest Unfairness due to Unresponsive Flows

Unfairness due toLong Propagation Delay:

Unfairness due to Long Propagation Delay FTP Source FTP Source Dest Dest 1 10

New MechanismNetwork Border Patrol (NBP):

New Mechanism Network Border Patrol (NBP) Our solution: Core-stateless congestion avoidance mechanism How it works:Uses adaptive feedback-based traffic policing to match the input and output rates Egress routers return explicit rate feedback to ingress router Ingress routers police traffic at input TCP-friendly rate control algorithm at ingress routers Feedback control algorithm detects congestion by monitoring the round trip time (RTT)

Network Border Patrol (NBP)Example Illustration:

FLOW 2 FLOW 1 6 Mbps 6 Mbps 4 Mbps 4 Mbps 4 Mbps 2 Mbps 4 Mbps 2 Mbps 4 Mbps 2 Mbps 2 Mbps 2 Mbps 2 Mbps 6 Mbps 6 Mbps 6 Mbps 6 Mbps 6 Mbps 6 Mbps 2 Mbps 2 Mbps 4 Mbps 4 Mbps 4.1 Mbps 4.1 Mbps 4 Mbps 4.1 Mbps Network Border Patrol (NBP) Example Illustration

Network Border Patrol (NBP)Example Illustration:

FLOW 2 FLOW 1 Network Border Patrol (NBP) Example Illustration

Network Border Patrol (NBP)Example Illustration:

FLOW 2 FLOW 1 Network Border Patrol (NBP) Example Illustration

Network Border Patrol (NBP)Example Illustration:

FLOW 2 FLOW 1 Network Border Patrol (NBP) Example Illustration

Network Border Patrol (NBP):

Network Border Patrol (NBP) NBP architectural components Ingress routers: police and rate control input traffic Egress routers: monitor traffic and return congestion feedback per flow base at edge routers Core routers: unchanged stateless at core routers End systems: unchanged NBP Internals Feedback control algorithm Rate control algorithm

NBP Architectural ComponentsIngress Router:

NBP Architectural Components Ingress Router Output port of an NBP ingress router

NBP Architectural ComponentsEgress Router:

NBP Architectural Components Egress Router Input port of an NBP egress router

NBP Feedback Control AlgorithmFeedback Control Packet Format:

NBP Feedback Control Algorithm Feedback Control Packet Format

Example of the Problem of Unfairness due to Unresponsive Flows:

Example of the Problem of Unfairness due to Unresponsive Flows Long queues at routers Unresponsive flow consumes most network bandwidth Current Internet

Example of NBP protection against unresponsive flows:

Example of NBP protection against unresponsive flows Short queues Fairer bandwidth allocation among competing flows Unresponsive flow’s packets are dropped at network input Internet with Network Border Patrol

Slide54:

Up to now application aggregated into a smaller number of flows flows hopefully are long lasting and not too dynamic control flows Future Current model may not be valid

Slide55:

Is what we do really what we need in future networks?

Where are we heading?:

Where are we heading?

BackgroundProcessing Capability Everywhere:

Background Processing Capability Everywhere Boundaries between game machines, home appliances, PCs, mobile phones are disappearing

Slide58:

Sony Playstation II hardware highlights 128 bits CPU, support for high definition 3D graphics 32 MB SRAM CD-ROM/DVD two USB ports one 1394 port one PCMCIA card slot Not just a game machine, it is a computer

Slide59:

Digital TV, Internet TV Watch TV programs on your PC and lap top Boundary between PCs/laptops and home appliances disappearing

Slide60:

Home appliances are being connected through a network Wired HomePNA (Home Phoneline Networking Alliance) Ethernet/Fast Ethernet IEEE 1394 Wireless Bluetooth IEEE 802.11

Slide61:

Wireless phone are becoming an information processing terminal I-mode by NTT Docomo

Slide62:

One dimension of network expansion home (home appliances, game machines) portable terminals (mobile phones, lab tops)

Background: Wearable Computers:

Background: Wearable Computers Wearable computers advanced forms (i.e., extremely small, light weight, integrated functions, evolvable, flexible) of wireless phone, hand held computers, video game terminals IC cards (digital cash, ID, bank cards, credit cards) googols CPUs in clothing Direct brain-computer interface

Slide64:

One dimension of network expansion small devices/chips on human users

BackgroundSensor Net and Autonomous Device Net:

Background Sensor Net and Autonomous Device Net Network of very small sensor devices device miniaturization (dust sensors, sensors in blood vessel, active bar codes, sensors in freeway） sensor: very simple function, simple communication capabilities

Slide66:

Dust sensors Float in the air to monitor environment conditions Sensor robots for Mars exploration Sensors for military applications Dump small sensors in the enemy territory Bio-degradable

Slide67:

Network of autonomous devices autonomous sensors mobile self-organizing to collectively perform one task flexible/evolvable (hardware and software) NASA, Mars exploration robots

Slide69:

One dimension of network expansion extremely small devices everywhere autonomous devices

Future Networks:

Future Networks Diverse network connecting human users and non-human devices everywhere (home appliances, mobile phones, wearable computers, small devices and autonomous devices) Extremely large scale network

Background: Applications:

Background: Applications Applications for fun email through mobile phone, i-mode (chat with friends) digital pets (e.g., pokemon, SONY robot dog) video games (e.g., interactive network games with high definition 3D graphics) players across the network (racing, fighting/war) # of games for wireless phones at Tokyo game show, fall ‘00

Future ApplicationsSome Examples:

Future Applications Some Examples Support for a large number of highly dynamic, highly customized small communities of users ad hoc creation of a net at a Michael Jackson’s concert; after the concert, ad hoc creation of a net in Rome among those who are pop music funs; etc., etc.

Slide73:

Extending human presence everywhere realistic and personalized/customized communication services transmitting human emotions/feelings (happy, sad, etc.) smells taste to appeal to human’s 5 senses

Slide74:

Virtual Society all human social functions done in a virtual society cyber-self (software entity representing you) conducts social functions (i.e., moves around on a net, meets other cyber-selves, make friends) conducts economic functions (i.e., conducts business (e.g., buy and sell goods)) conducts biological functions (i.e., mates, produces off springs)

Slide75:

Highly customized and personalized applications Future applications are NOT: digital libraries remote operation of medicine distance learning Federal information services

Future Networks:

Future Networks Extremely large scale, diverse network connecting human users and various devices Highly dynamic, diverse, realistic, and personalized applications

Slide77:

“application aggregated into a small number of flows” model may not be valid private highly dynamic diverse traffic characteristics large scale, but may have locality I don’t need to control a refrigerator in Rome hot spots

What do we need to do?:

What do we need to do?

Slide79:

Are Higher speed and better QoS important? Not sufficient to handle scalability diversity (of users, of applications) dynamic environments realistic communications customizability/personalization adaptability/flexibility/evolvability

Applications:

Applications Identify promising applications Study application requirements, and design a net based on them

Slide81:

Example: Internet TV does not necessary need higher speed and better QoS popular usage will be: set top box at home screens TV programs/shows based on user preference stores them in a storage device at home view them later when users have time no strict QoS and high speed requirement on networks

Slide82:

Example QoS (location accuracy, reliable communication) may not be important for some applications (e.g., for sensors) world-wide roaming may not be important for highly personalized friends-chatting-with-friends applications

Diverse Applications over a Small Set of Infrastructures:

Diverse Applications over a Small Set of Infrastructures Need to support a huge number of highly personalized diverse applications Interface one-size fits all underlying net to diverse applications where and how? middleware becomes important

Slide84:

Middleware needs to map application level QoS onto network level QoS QoS aware middleware Application, middleware, network mechanisms helping each other to achieve QoS

Distributed/Local Architecture:

Distributed/Local Architecture Pursue 100% distributed network architecture no central or coordinating entities assume autonomous entities build a large scale system from a simple localized algorithms/behaviors/designs example: no directory service discover when needed by asking “friends”

Slide86:

Accept overhead we gain other features (self-organization and -coordination, flexibility, adaptability, service emergence and evolution, etc.)

Example: Bio-Networking:

Example: Bio-Networking Observation: large scale biological systems scale, adapt, and survive Apply biological concepts/mechanisms to future Internet applications emergent behavior (out of simple behaviors) lifecycle (food/energy, reproduction, death) evolution through diversity and natural selection etc.

Slide88:

Bio-Net individuals = cyber-entities (objects/agents) abstraction of various system components (users, resources, service components) autonomous with simple behaviors e.g., replication, reproduction, migration, energy exchange, relationship establishment, pheromone emission, death makes its own decision, according to its own behavioral policy

Slide89:

Cyber-entity example behaviors energy exchange gain energy from a cyber-entity (e.g., a user) in exchange for performing a service expend energy to receive service from other cyber-entities (e.g., to use network/computing resources) can be used as a natural selection mechanism evolution

Slide90:

relationship establishment a cyber-entity knows something (e.g., name, address, service type) about another cyber-entity amount of information strength of relationship service emergence

Vision:

Vision No central or coordinating entity exists. A large number of CEs (created by millions of millions of Internet users), autonomously moving/replicating, CEs contacting other CEs providing related services, making relationship, diverse behavior policies getting created, good behaviors survive, bad ones die, making system flexible, adaptable and evolvable

Economic and Social Implications of IT:

Economic and Social Implications of IT Social, economic and workforce issues related to information technology legal and regulatory constraints tax treatment of investment in information technologies workforce skilled in the use of information technologies

Slide93:

Are we doing something good to ourselves? Email is very convenient, but Flooded with junk emails, and cannot find time to do what I want to do Wireless phone is very convenient, but It is very annoying in a train It is a health risk to those who wear heart pace maker.

What kind of techniques and mechanisms do we need? (near term):

What kind of techniques and mechanisms do we need? (near term)

Examples of Techniques We Need:

Examples of Techniques We Need Flexible/modular network architecture easy to add more (or reallocate existing) resources (hard/soft resources such as CPU, buffer, finer level resource) Mechanisms to help processing within a network not “transmit fast, process at end systems” process data on transit in a network consider link/network as buffer a link with data processing capability?

Slide96:

Mechanisms to build a large scale system from a simple localized algorithms, behaviors, designs self-organization and coordination mechanisms service finding mechanisms resource allocation/management mechanisms

Slide97:

Mechanisms to anticipate/predict and prepare to give users illusion of a dedicated high speed, good quality network Understanding large scale networks network modeling and simulation

Examples from Soft/Middleware:

Examples from Soft/Middleware Active/smart software software that updates itself, monitors its progress toward a particular goal, discovers/downloads a new capability needed for the task at hand Self-organization, -coordination techniques

Summary:

Summary

Slide100:

Tremendous Internet growth We’ve been busy doing near term research We probably need to take a moment and think what future looks like and what we need to do

Slide101:

Thanks!

Bio-Networking Architecture:

Bio-Networking Architecture

Motivation:

Motivation Network services/applications need to be scalable adaptable (to heterogeneous/dynamic conditions) survivable and available simple/easy to design and maintain Networks need to have built-in mechanisms to provide these features large nets; beyond one’s capability to design

Slide104:

Observation: large scale biological systems scale, adapt, and survive Apply biological concepts/mechanisms to future Internet applications emergent behavior (out of simple behaviors) lifecycle (food/energy, reproduction, death) evolution through diversity and natural selection etc.

Emergent Behavior:

Emergent Behavior Biological systems useful group behavior emerges from autonomous local interaction of individuals with simple behaviors

Slide106:

Slide107:

Slide108:

relationship establishment a cyber-entity knows something (e.g., name, address, service type) about another cyber-entity amount of information strength of relationship

Slide109:

relationship can be used to group cyber-entities collectively providing a service application constructed from a collection of cyber-entities e.g., a web server (application) from a collection of web pages (cyber-entities)

Evolution and Adaptation:

Evolution and Adaptation Biological systems the biological system adjusts itself for environmental changes of long-term and short-term key components diversity from mutations and crossovers during replication/reproduction natural selection keeps entities with beneficial features alive and increase reproduction probability

Slide111:

Bio-Net cyber-entities (CEs) evolve, adapt, and localize through diversity and natural selection diversity A CE behavior can be implemented by a number of algorithms/policies human designers can introduce diversity in CE behaviors CEs replicate/reproduce with mutation/crossover in behavior policies

Slide112:

natural selection (energy as a natural selection mechanism) death from energy starvation tendency to replicate/reproduce from energy abundance

Slide113:

Sensor CE Image analyzer CE Stereo CE Some CE Video camera CE Rock MP3 CE Latin MP3 CE Example Application

Slide114:

Sensor CE Image analyzer CE Some CE Video camera CE Rock MP3 CE Latin MP3 CE Example Application

Slide115:

Sensor CE Image analyzer CE Some CE Video camera CE Rock MP3 CE Latin MP3 CE Example Application

Slide116:

Sensor CE Image analyzer CE Stereo CE Some CE Video camera CE Rock MP3 CE Latin MP3 CE Example Application

Example CE BehaviorAdaptation Simulation:

Example CE Behavior Adaptation Simulation Cyber-entity behaviors implemented replication, death, migration

Cyber-Entity Behaviors:

Cyber-Entity Behaviors Replication If current energy level > aggressiveness, then create a new entity of same type Death if current energy level = 0, then, die Migration migrate towards source of energy (user requesting service) avoid coexisting on a node with same entity