Docstoc

TCP peril

Document Sample
TCP peril Powered By Docstoc
					미래 application 들을 위한 network
            solutions에 대한 연구

          March 23, 2004
          Younghee Lee




                                               1
              Prof. Younghee Lee   한국정보통신대학교
Content
 Requirements      for future applications
  – NGI
  – Ubiquitous computing
 Problems  of today’s Internet
 Possible solutions
  –   MPLS
  –   Active & programmable network
  –   Overlay network
  –   Ad hoc network, Sensor network
  –   (knowledge plane)
 Research    activities of cnlab ICU
                                                             2
                            Prof. Younghee Lee   한국정보통신대학교
Requirements for future the Internet applications
  NGI
   – Increased Capability
      » Advanced end-to-end networking technologies:
          Reliability, Robustness, Security, QoS/differentiation of service
           (including multicast and video), Network management (Including
           allocation and sharing of bandwidth)
   – Increased Capacity
      » The ―100x‖ testbed — at speeds 100 times faster end to end than
        today‘s Internet.
   – Applications
      » Collaboration technologies, Digital libraries, Distributed computing,
        Privacy and security, Remote operation and simulation


                                                                                3
                                  Prof. Younghee Lee          한국정보통신대학교
Requirements for future the Internet applications
   Networking for Ubiquitous computing (IBM)
    – Plug-and-play networking
    – Requires ―smarter‖ infrastructure
         » Self configuration
         » Auto-discovery and Service access
    –   Proximity based connectivity
    –   Hidden computing
    –   Spontaneous networking
    –   Security and Privacy
    –   Access and connectivity rules




                                                                          4
                                         Prof. Younghee Lee   한국정보통신대학교
Requirements for future the Internet applications
    Architecture for pervasive computing system
      – Networked Apps, API
         » Power conscious Apps
         » HW and middleware aware
         » Disconnected model when possible
     – Middleware/Networking Stacks
         » Utilizing existing stacks if possible
         » Lightweight networking for peers
         » Complexity pushed to infrastructure for Internet access
     – Radio/BB/MAC
         »   Integrated RF design
         »   Low power transmitter (1 mW)
         »   Power conscious MAC
         »   Encryption
         »   Ubiquitous system interface



                                                                                 5
                                            Prof. Younghee Lee       한국정보통신대학교
Requirements for future the Internet applications
    2-5 years later
      – Environment
          » Smart Spaces, Internet Appliances, Things-that-think, Car, Home Networks, Body-on-
            the-Net
      – Technology
          » Intergrated/embedded Networking (low cost, Low power), "Lightweight" IP and
            Networking Services, Spontaneous Networking, Wireless, Universal connectivity
    The future of the Internet is not multimedia(only).
      – The future of QoS networks is cloudy
      – Primary driver for advanced networking?
          » The future multimedia applications? or
          » computer to computer data networking
    Challenges in nomadicity:
      – Location independence, Device independence, Widespread access, Security,
        Adaptability to new technologies, Friendly interface, Partitioning functionality into
        co-operating software entities


                                                                                                 6
                                           Prof. Younghee Lee             한국정보통신대학교
Requirements for future the Internet applications
   Different applications needs different security services
    – Confidentiality, Integral, availability, non-repudiation, Access control,
      Authentication
    – lacking : authorization and quality of service
   Four concrete application areas for further bandwidth usage:
    – Real-time synchronization: User will have data and media on various devices and
      in various central server and decentral (think your P2P storage of movies) places,
      and user want to have access to this data and media all the time everywhere.
    – Media usage:If it can take only 10 seconds for movie download, people will use that
      bandwidth. Then they will sample movies and TV and download more. User will
      share with friends sending a full movie via email like mp3 files?
    – Life Storage: By far the biggest driver will be the recording, transmission and
      storage of whole life. (Data, Information, knowledge, wisdom)
    – Gaming: When you get real-time 3D worlds inhabited by thousands of avatars
      interacting with each other, and your local gaming environment always needs to be
      on top of all the developments in this world, then your bandwidth usage will
      skyrocket. NxN multicast
                                                                                        7
                                      Prof. Younghee Lee           한국정보통신대학교
Questions
   Network service 측면:
    – 어떻게 저 많은 서비스 요구사항을 네트워크가 만족시킬 수 있는지?
    – 현재 망 차원에서는 불가능?
   Network control 측면:
    – 어떻게 저 다양한 응용서비스가 네트워크를 control 하면서 각자의
      구미에 맞게끔 네트워크를 이용할 수 있을지?
    – No control plane for application in the Internet(best effort service case)




                                                                                   8
                                   Prof. Younghee Lee          한국정보통신대학교
Problems of today’s Internet
   Internet design principles
    – Internet Architecture : Cerf and Kahn‘s internetworking
      principles:
       » minimalism, autonomy - no internal changes required to interconnect
         networks
       » best effort service model
       » stateless routers
       » decentralized control
         Big differences with connection oriented
         telecommunication networks (PSTN, PSDN, ATM,…)



                                                                               9
                                 Prof. Younghee Lee        한국정보통신대학교
Problems of today’s Internet
   Internet design principles
    – End-to-End Argument
       » If the application can implement a functionality correctly, implement it
         a lower layer only as a performance enhancement
             Application has more information about the data and the semantic
              of the service it requires (e.g., can check only at the end of each
              data unit)
             A lower layer has more information about constraints in data
              transmission (e.g., packet size, error rate)
       » Rule of Thumb
             Implementing a functionality at a lower level should have minimum
              performance impact on the application that do not use the
              functionality
       » What About Other Services?: Multicast? Quality of Service (QoS)?


                                                                                    10
                                   Prof. Younghee Lee          한국정보통신대학교
        Why Is It Not Happening?
   Network QoS model is too primitive.
                                                                Distributed
     – Large gap between network and                            Simulation
       application QOS                                                  Distance
     – Too low level; hard to use                                       Learning
   Applications have insufficient                  User       Games                    User
    information about the network to make                                 Video
                                                                       Conferencing
    informed decisions.
     – Am I using a modem or a gigabit
       Ethernet?
                                                                                          No
     – Where can I get more bandwidth            Too Complex      No Control          Information
   Service providers have little control
    over how their traffic is handled.
      – No customization
   Implication to active network, overlay
    network, ad hoc network?
   Knowledge plane?

                                                                                               11
                                         Prof. Younghee Lee             한국정보통신대학교
Problems of the Global Internet
   Problems and requirements
     – Various Internet attackers: spam e-mail,..
         » Need protect users and network itself from attacker
     – ISP Service differentiation: QoS
     – Third party‘s involvement
         » 정부, ISP 등이 위해정보 차단, 세금징수 등..
     – Multiway communication
     – Firewall in the network, traffic filters, NAT for address space management
     – (Congestion control, incentive)
 How    can we solve all these problems or requirements of the Global
    Internet ?
     – Location?
     – Labeling for packet discrimination?…
     – How to improve and evolve current Internet respecting e2e argument principle?

                                                                                    12
                                           Prof. Younghee Lee    한국정보통신대학교
Network processor
    Today’s gateways and backbone routers can never be
    fast enough
     – Cannot keep up with fiber capacity
     – Cheap, monolithic “superprocessor” : Intel IXP nnnn
        » Replacing rack-mount routers
        » Also processing higher layer protocol
        » QoS, encryption




                                                                       13
                                      Prof. Younghee Lee   한국정보통신대학교
MPLS
 QoS   routing?
   – Bandwidth?, Delay?
   – Delay-constrained least cost (NP-complete)
          IntServ
 DiffServ,
 Stateless or stateful ?
   – Challenge: features of stateful solutions, but at the cost of
     stateless solutions
   – Can MPLS be a candidate ?
      » Differentiate flows for optimum performance and services
           Push complexity of control plane to data plane

      » Positive. But what about IP network design principle? : Stateless


                                                                              14
                                    Prof. Younghee Lee            한국정보통신대학교
Programmable Platforms
   Stateful solutions need a complex control plane
    – Control plane: difficult to develop and debug
    – open flexible control plane
   Open programmable interface (API)
    – user, network node, third party : resource manipulate or
     reprogram
    – open signaling: IEEE 1520
   Related standard
    – IETF General Switch Management Protocol(GSMP)
    – Forwarding and Control Element Separation (ForCES)
    – The Multiservice Switching Forum (MSF)

                                                                       15
                                 Prof. Younghee Lee        한국정보통신대학교
Active Networks
   Various active network research(mostly funded by DARPA)
    – ANTS/PAN ( MIT), SwitchWare (Upenn), Liquid Software (Arizona),
      NetScript (Columbia), Janos (Utah), ANTS/Detour (Washington),
      OpenetLab (Nortel), CANES (Georgia Tech), Genesis (Columbia), Panda
      (UCLA), Smart Packets (BBN), DARWIN (CMU), Active Networks and
      Novel Network Management Technology (GE), ABLE (Bell Labs)
    – Very active until 2000
      » About 6 projects are active now in US
      » Activeware (MIT) Liquid Software (U. Arizona) Scout Operating System (U. Arizona)
        Spin Operating System (U. Washington) Switch Ware Project (Upenn NOW Network
        of Workstations (U. Berkeley)
   FAIN(Future Active IP Networks)
    – R&D project under the Information Society Technologies (IST) program : 3
      years from 2000 funded by Commission of the EU
    – Various European countries, Hitachi, Upenn
    – open, flexible, programmable and dependable (reliable, secure, and
      manageable) network architecture based on novel active node concepts.
                                                                                            16
                                      Prof. Younghee Lee            한국정보통신대학교
Active Networks
 – ―programmability into the network‖,
 – ―new services are introduced fast‖.
 – Problems
  » What is the killer application? Incentive?
  » Need of processing power
 – end to end argument point of view
  » contradict the end-to-end principle: a function or service should be
    carried out within a network layer only if it is needed by all clients of that
    layer
  » consonant with end-to-end arguments: programmability may allow a
    network client to implement precisely the service it needs, an outcome




                                                                                     17
                                 Prof. Younghee Lee            한국정보통신대학교
 Active networks for something
   Execution at right place;
    – Something would prefer to be executed at intermediate node rather than
      end node
 Characteristics       of functions
    – closely related with network control or node data manipulation
        » Congestion, multicast, QoS, sensor node,…
    – reactive on right time at right place
        » Adaptive to network changes or context changes
        » For pervasive computing
             Context => event => service

                 – How to describe service?: service description
                 – How to discover optimum service?: service discovery
                 – How to execute them on right time at right place?: agent
    – application specific and temporal only to certain application
        » If it‘s common to every application, we don‘t need active networking
                                                                                          18
                                          Prof. Younghee Lee                  한국정보통신대학교
 Active networks for something(1)
 Active   networking for the GRID
   – Active P2P Grid architecture
   – Self-adapting, self-configurable, self manageable grids
 Active   overlay network
   – Application Level Active Networks (ALAN)
 Active sensor network
   – Adaptive to network situation, decided by application
   – Active routing for ad-hoc network: Seamless integration
 Programmable network management
   – Applying Active networks to Network management
       » Smart Packets(BBN)



                                                                           19
                                   Prof. Younghee Lee          한국정보통신대학교
Active networks for something(2)
   Active Networking in Pervasive Computing
     –   situation(context) aware, dynamic, adaptive,….
     –   Mobility discovery
     –   Pan-network server service
     –   Agent: execution on behalf of application at better place
   Active Networking for OPES
     – OPES: Services deployed at application level intermediaries i in the network to
       transform filter content
           » Caching, virus scanning, language translation, … , …
     – Active node to execute the code for specific application
           » Message containing the code or rule set
                video transcoding, virus filter and so on...

   Active networking for knowledge plane concept
     – Knowledge based network control for effective network
     – Agent execution for various applications
     – Information gathering, knowledge discovery, …
                                                                                         20
                                               Prof. Younghee Lee     한국정보통신대학교
 Overlay Network
 Motivations
  – Changes in the network happen very slowly
  – Why?: Network services are end-to-end
  – Proposed changes that haven‘t happened yet:
       » Congestion (RED ‗93); More Addresses (IPv6 ‗91), Security (IPSEC ‗93); Multi-
         point (IP multicast ‗90)
 Anisolated virtual network deployed over an existing
 network
  – Composed of Hosts, Routers, Tunnels
  – IP service: e2e datagram service
  – Multicast, QoS services need stateful protocols only for control
    state over IP networks => e2e edge/overlay service
  – Application level intermediaries
                                                                                         21
                                     Prof. Younghee Lee           한국정보통신대학교
 Overlay Network
 New   service deployment without network updates
   – Performance drawback compared to the case with network updates
 Potential   Benefits
   – Easier to deploy
        » only requires adding software to end hosts
   – Potentially simplifies support for higher level functionality
        » leverage computation and storage of end systems
             e.g., packet buffering, transcoding of media streams, ACK
              aggregation
        » leverage solutions for unicast congestion control and reliability




                                                                              22
                                     Prof. Younghee Lee          한국정보통신대학교
    Overlay Network: applications
   Applications
    –   Multicast
    –   Quality of Service
    –   Mobility
    –   Addressing: 6bone, IP-NL ; enhanced NAT
    –   Security
    –   Web caching, CDN, P2P
         » Related IETF activities
               Web Replication and Caching (WREC)
                  – Taxonomy, requirements
               Content Delivery Internetworking (CDI)
                  – Settlements, SLAs, property rights
               Web Intermediaries (WEBI)
                  – Content Invalidation Protocol
               Open Pluggable Edge Services (OPES)
                  – Rules-based invocation of proxylet services
                                                                          23
                                   Prof. Younghee Lee         한국정보통신대학교
    Overlay multicast: (Overcast)
   Scalable, efficient, and reliable distribution of high quality video
   Large groups ~ millions of nodes
    – Typical application: content distribution
   Designed for throughput intensive content delivery
    – Streaming, file distribution
    – Not good for gaming application: latency problem
   Server based infrastructure
   ICU :
    – 1 to N, N to N multicast for streaming service
    – High performance forwarding engine in kernel level




                                                                           24
                                  Prof. Younghee Lee       한국정보통신대학교
 Knowledge Plane
 Concern      over risks of increased reliance on networks
     – The role of the network is growing more quickly than our ability to manage
     – Network-centric warfare has promise and peril
     – The civilian economy is alternately helped and hurt by the Internet
 Key    Idea: The Internet Knowledge Plane as a basis for making
    progress in cognition while exploring a new vision for network
    architecture
     – New ―collective cognitive‖ mechanisms for supporting cooperation and
       learning
     – A coherent management infrastructure for the Internet that does not
       compromise its strengths ;e2e
     – Additional military benefits: quick deployment, more effective networks, and
       reduced reliance on human experts
   초기개념 형성단계
   다양한 응용서비스가 필요한 지식정보를 공유: 응용별로 망 구성을 위한 별도의
    정보 획득 및 조치 불 필요
                                                                                    25
                                     Prof. Younghee Lee         한국정보통신대학교
Knowledge plane:
                     THE KNOWLEDGE PLANE

                                     K-Application “Why?”:
                              Network fault detection, isolation, and repair

                             K-Base                                 Models

                             Inference                             Models of
                               rules,                         Internet structure,
                            diagnostic                            application
                            procedures                             behavior,
                                                                 requirements
   Perception                                                                        Action
    Sensors                                                                         Actuators



                •Departures from expectation
       E
                • Departures from design
                                    • Element failures
                                    • Misconfiguration                                          E

                                    • Attacks
                    E
                                                                E




                                                                                                    26
                                                Prof. Younghee Lee                      한국정보통신대학교
Knowledge plane:Technology Foundations

                              Algorithmic game theory
                                                                   Bayes belief nets, machine
   Domain-specific                                                 learning, genetic algorithms,
   languages                                                       neural networks, expert
                                                                   systems




  RKF, DAML,
  Knowledge                                M    P   K
  Representation,                                                  Distributed Hash Tables
  dimensionality                                                   (DHTs)
  reduction




           Active Networks, Sensor Nets,                        DASADA, NMS
           CoABS, various overlay
           networks

                                                                                                   27
                                           Prof. Younghee Lee               한국정보통신대학교
Knowledge plane:Technology Foundations




                                                     28
                    Prof. Younghee Lee   한국정보통신대학교
Knowledge plane: summary
  An   net that builds itself using high-level specification.

  Very    different net from the Internet.
    – We might experiment with knowledge overlays

  What     is different?
    – Edge-involvement.
          » Visibility of ―application-level‖ behavior.
    –   Global perspective.
    –   Compositional structure.
    –   Unified approach.
    –   Cognitive framework




                                                                               29
                                              Prof. Younghee Lee   한국정보통신대학교
Ad hoc network: application
   Military environments: was motivation & strong candidate
     – soldiers, tanks, planes
           » Need mobility, avoid SPF, rapidly deployable, Multi-hop to reach to person outside of LOS(line of
             sight), when existing infrastructure is unavailable
     – Survivable Radio Network(SURAN), Global Mobile(GloMo) Information System
   Civilian environments
     –   taxi cab network, automobile communications(Cellular + ad hoc+..)
     –   Meetings/conferences
     –   sports stadiums, super market, Hotel…
     –   boats, small aircraft
   Emergency operations
     – search-and-rescue
     – policing and fire fighting
   Personal area networking
     – cell phone, laptop, head phone, wrist watch, multimedia devices
     – Wearable computing

                                                                                                                 30
                                               Prof. Younghee Lee                    한국정보통신대학교
Ad hoc network
 MANET   nodes
  – End system and also Network nodes
     » Discussion: Aspect of “End to End Arguments” in MANET?
  – With wireless mobile host
  – May need multiple hops to reach a destination




                                                                            31
                                    Prof. Younghee Lee          한국정보통신대학교
 Sensor Network
 Applications    of sensor network
   –   Home network for pervasive computing
   –   Habitat monitoring
   –   Environmental observation and forecasting systems: Columbia River Estuary
   –   Smart Dust
   –   Biomedical sensors
   –   Military applications




                                                                               32
                                    Prof. Younghee Lee       한국정보통신대학교
Classifications of Sensor Nets
 Sensor    position
  – Static (Habitat, CORIE, Biomedical)
  – Mobile (Smart Dust, Biomedical)
 Goal-driven
  –   Monitoring: Real-time/Not-real-time (Habitat, Smart Dust)
  –   Forecasting (CORIE)
  –   Function substitution (Biomedical)
  –   …
 Communication        medium
  – Radio Frequency (Habitat, CORIE, Biomedical)
  – Light (Smart Dust)

                                                                  33
                          Prof. Younghee Lee     한국정보통신대학교
     Common Challenging Issues
   Limited computation and data storage
    – Sensor design (Multi-objective sensors), Cooperation among sensors
    – Data aggregation and interpretation
   Low power consumption
   Wireless communication
    – Medium, ad hoc vs. infrastructure, topology and routing
   Data-related issues
   – Trade-off between latency and energy: reactive? proactive?
   – Data representation: Raw/Compressed data
   – Error calibration: No access to real values, Inferred from other sensors
 Continuous operation: Long-term data collection
   – Renewable power source.: Solar energy, Mechanical vibrations, Radio-
      Frequency inductance, Infrared inductance
   Inaccessibility – network adjustment and retasking
   Robustness and fault tolerance
                                                                                34
                                     Prof. Younghee Lee          한국정보통신대학교
Uncertain   Conclusion

 Need many thing between applications and very
 high speed networks
  – Pay too much attention only to HSN?
  – Intermediaries: Middleware
 Interimsolution: overlay network?
 Ultimate solution?
  – Knowledge plane?
  – Totally new global network?
 Solutions   for local environment?
  – Sensor network, ad-hoc network, WPAN,…

                                                          35
                         Prof. Younghee Lee   한국정보통신대학교
 Computer Network Lab.
 People
  – 7 Ph.d students, 4 Ms Students
 Research
  – Network Supports for Pervasive Computing In Home
    Networking environments: making home more comfortable,
    safe and convenient, controlling devices automatically
    without user’s knowledge
     » Pervasive Network Access
         Zero-configuration performed over entire networks of nodes
         Mobility management: adaptive mobility

     » Context aware semantic service discovery
           Automatic service discovery with minimized user’s intervention

                                                                             36
                                 Prof. Younghee Lee       한국정보통신대학교
Computer Network Lab.
 Research
  – Active networking: making the network intelligent and
    programmable for high quality Internet services
      » Congestion control, multicast, QoS, sensor network node,…
      » Reactive on right time at right place
  – Overlay Network: making the end node computers working
    like network nodes           immediate new network service
      » Overlay multicast: Split -join
      » Programmable overlay
  – Ad hoc network: making the computer nodes to construct
    the network by themselves
      » Ad hoc routing: Proactive-reactive Hybrid type
      » Address auto-configuration
                                                               37
                           Prof. Younghee Lee   한국정보통신대학교
Computer Network Lab.
 Research direction
  – Adaptive networks
    » Self configuration: zero configuration
         Mobile devices, ad hoc devices,…

    » Dynamically adapt to the requirements of applications and situation
      changes
  – Service discovery
    » Semantic service discovery: Currently Home network environment
        Inexact matching

        Interworking between existing middleware ;Jini, Havi, UPnP…

    » Extend to Global network environment including mobile network
        OSGI


                                                                            38
                               Prof. Younghee Lee         한국정보통신대학교
       Semantic service discovery
                                                                                    class

                                     Entity                                         subclass

                                                                                    property




                                              Primitive       State      Control
Attribute     Device       Service
                                              Service        Variable   Interface




     Ontologies in home environment
        – Advantage of our ontology structure
            Low complexity
            Easily define relation between device and service
            Enabling the composition of services and device attributes based
             query message

                                                                                               39
                                              Prof. Younghee Lee          한국정보통신대학교
  Semantic service discovery
 Ontology structure
   Device ontology
    A smallest physical unit of providing a service
  Service ontology
    Primitive service composition, and primitive service and device attribute
     composition
  Primitive service ontology
    A smallest logical unit of providing service
    A mediator between device and service
  Attribute ontology
    Device attributes
    Represent device attribute efficiently
  State Variable / Control Interface ontology
    models state of primitive services with state variables and control primitive
     service through control interfaces
                                                                                     40
                                     Prof. Younghee Lee              한국정보통신대학교
Implementation
    System Model
      Jini-based client / service model


                                             Lookup
                            3. reasoning &
                               evaluation
                                             Service                2.Download
        Ontology                             Reasoning           device-description
                                               Engine                   file             Device-description
                                             Evaluator                                       Repository
                                              Matching
                                              Manager
                                                         Service
                   4.Discovery&Lookup                     Proxy1.Discovery & Join



                                        5.Receive
                      Service
                       Proxy

             Client                            6.Use                           Service




                                                                                                              41
                                              Prof. Younghee Lee                      한국정보통신대학교
    Implementation
   Architecture of Extended Lookup Service

           Lookup Service

                                     Registry                                                               Device description
                                                                                                               Repository



                               Inferencing Engine                                                            Device/Service
                                                                                                              Description
            Request   Reply                                                                                    Ontology


                              Evaluator                  Dynamic-value              Request dynamic value
                                                                                                             Location Server
                                                           Extractor                  Reply dynamic value
                                                                                                                    .

                                                                                                                 Device
                        Request of     Result of evaluation                                                         .
                                                                                                                    .
                        evaluation
                                                                 Query
     Service
     register
                                Matching Manager               Interpreter
                                                 Service request             Service request
                                                        message              message                                    Control
                                                                                                                        Data
       Service                                                     Client

                                                                                                                                  42
                                                          Prof. Younghee Lee                                한국정보통신대학교
NGIS middleware
 NGIS   미들웨어 구조

                NGIS Applications: Internet TV, Multimedia Comm.


                            NGIS Middleware Architecture

    NGIS 네트워킹 미들웨어                                        메타데이타 변환 엔진
    N to N Multicast           1 to N Multicast           MPEG-7 Ontology
    - Topology Management      - Bandwidth Acquisition    - Ontology Management

           Fast Packet Forwarding Engine                  Transformation Engine

                        IXP 기반의 고성능 인터넷 정합장치




                                                                                    43
                                     Prof. Younghee Lee                 한국정보통신대학교
Networking Middleware
 1-to-N   Multicast
  – Target: High-quality multimedia streaming
  – Requirement: Bandwidth Stability
  – Approach: Split & Combine
                 Level-0 : Sender
                            Level-1



                                      Level-2




                                                             44
                            Prof. Younghee Lee   한국정보통신대학교
Networking Middleware
   Fast Packet Forwarding Engine
    – Overlay Multicast 패킷을 적은 Latency로 Forwarding
    – Forwarding Engine: Kernel 영역에 위치
    – buffering의 최소화




                                                            45
                          Prof. Younghee Lee    한국정보통신대학교
Networking Middleware
   Fast Packet Forwarding Engine
    – Protocol Update Engine: 현재 등록된 Protocol의 정보 수정
    – Protocol Interface: Enhanced Socket Interface for Overlay Multicast




                                                                            46
                               Prof. Younghee Lee         한국정보통신대학교

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:3
posted:12/2/2011
language:Korean
pages:46