Document Sample
TCP Powered By Docstoc
					Computer Networks
              Chap.        6 QoS and Multicast


               교수 이영희

                                          1   1
           Prof. Younghee Lee
     Why Is It Not Happening?
   Network QoS model is too
    primitive.                                             Distributed
     – Large gap between                                   Simulation
       network and application                                      Distance
       QOS                                                Games
     – Too low level; hard to use
                                        User                         Video
   Applications have
    insufficient information
    about the network to make                                                        No
                                     Too Complex              No Control         Information
    informed decisions.
     – Am I using a modem or a
       gigabit Ethernet?
     – Where can I get more
   Service providers have little
    control over how their traffic
    is handled.
     – No customization              Prof. Younghee Lee
                                                                                  2      2
      Internet Traffic: Inelastic Traffic
   Traffic which does not easily adapt to changes in delay and
    throughput across the Internet: real time traffic
   Requirements
    – Throughput: A minimum throughput value
    – Delay: delay sensitive: ex) stock trading
    – Jitter: delay variation: teleconference require a reasonable upper
      bound on jitter
    – Packet loss: the amount of packet loss sustainable
   New requirements
    – some means to give preferential treatment to applications with more
      demanding requirements.
        » Application need to be able to state their requirements, either ahead of
          time or on the fly: stating ahead of time is preferable(negotiating)
        » elastic traffic must still be supported in times of congestion, inelastic traffic will
          continue to supply a high load, and elastic traffic will be crowded off the Internet

                                                                                         3         3
                                          Prof. Younghee Lee
     Admission Control
   Admission control  deciding when the addition of new people would
    result in reduction of utility
     – Basically avoids overload
   Problem: It requires the concurrence of all the nodes situated along the
     – need global information
         » not practical:
              diverse requirement of MM application. Broadcasting information in real time:
              Large latency at gigabit speed: obsolete, inconsistent
   Node Admission Control(NAC): the process of deciding whether a node
    can admit a new connection
   Issues in admission control
     – the parameter specified by the user may not be correct. Conservative? -> low
     – admit new traffic taking into consideration the statistical smoothing effects?
     – Traffic existing in the network can be calculated from the user specified
       characteristics or be based on actual measurement?
         » Admission control based on measurement: achieve high utilization
                observation interval: window -> window size? ;issue
                decision made based on the past behavior + measured data
                                                                                   4       4
                                       Prof. Younghee Lee
    How Things Fit Together

 Routing                                                                              RSVP
                        Routing           RSVP
Messages                                                                             messages

                                                                    Control Plane

                                                                        Data Plane
             Forwarding Table           Per Flow QoS Table

   Data In
             Route Lookup         Classifier            Scheduler                    Data Out

                                                                                       5    5
                                  Prof. Younghee Lee
      ISA Service level
   a number of general categories of service are provided
     – Guaranteed: hard real-time (“real-time” applications)
        » For intolerant and rigid applications
        » Fixed guarantee, network meets commitment as long as clients send at match
          traffic agreement
    – Control load: soft real-time (“tolerant and adaptive” applications)
        » network to client: similar performance as an unloaded best-effort network
        » client to network: the session does not send more than it specifies
        » Two components
              If conditions do not change, commit to current service

              If conditions change, take steps to deliver consistent performance (help apps
                minimize playback delay)
              Implicit assumption – network does not change much over time

        » video: adaptive by dropping a frame or delaying the output stream slightly
        » voice: adaptive by adjusting silent periods.
    – Best effort: (“elastic” applications)
                                                                                     6         6
                                        Prof. Younghee Lee
    Packet Scheduling

   Network meet promises by scheduling
    – Queuing at router
    – Token bucket filter to characterize traffic
       Possible Uses
          1. Shaping
                Delay packets from entering point
          2. Policing
                May drop packets that arrive without token
                or
                Marking: Let all pkts pass through with marking

                                                                   7   7
                              Prof. Younghee Lee
      Token bucket
   Token bucket traffic scheme
    – A token replenishment rate R:
      specifies the continually
      sustainable data rate
    – A bucket size B: specifies the
      amount by which the data rate can
      exceed R for short period of time
    – During any time period T, the
      amount of data sent cannot exceed
      RT + B
    – Bucket: represent a counter that
      indicates the allowable number of
        » Bucket fills with octet tokens at the
          rate of R

                                                            8   8
                                       Prof. Younghee Lee
      ISA Services
   Token bucket traffic scheme
    – two counters: token counter, counter to
      implement the timer
    – major advantage: simplicity
    – P :peak rate, A :average rate, R :token
    – A peak rate limiting spacer is implicit: R
    – P > R > A:
    – Maximum burst size: b’
        » b(t)= B + (R-P) x t : assuming token
           is full at the beginning
        » 0 = B + (R-P) x t -> t = B/(P-R) ->
           b’ = Pt = B/(1- R / P)

                                                            9   9
                                       Prof. Younghee Lee
      Scheduling: Queuing discipline
   Drawbacks to the FIFO queuing discipline
    – No special treatment to packets from flow of higher priority such as delay
      sensitive traffic flow
    – larger average delay per packet than if the shorter packets were
      transmitted before the longer packet. Flows of larger packets get better
    – A greedy TCP connection can crowd out altruistic connections. Ex) RTO
      backoff in congestion
   Fair Queuing
    – A router maintains multiple queues at each output port
    – round robin, skip over empty queues. Load-balancing, protect greedy..

                                                                      10    10
                                  Prof. Younghee Lee
     Processor sharing
   Drawbacks of fair queuing scheme
     – Short packet are penalized
          » one packet per cycle
   Processor sharing
     – transmit one bit from each queue on each round
          » not practical to implement
          »Fi : value of R(t) when packet i in queue alpha ends transmission

                    d               1
       R ' (t )       R (t )                 ; rate of eac h round
                    dt          max[1, N (t )]
       Fi  Si  P ;

       Si  max[Fi1 , R ( i )]

   BRFQ
     – solve the problem in FQ: “short packet are penalized”
          » BRFQ: uses packet length as well as flow identification
     – designed to emulate PS
     – rule: whenever a packet finishes transmission, the next packet sent is the one with the
       smallest value of Fi
     – Good approximation to the performance of PS. Figure 11.10

                                                                                     11      11
                                             Prof. Younghee Lee
 Processor sharing



                                   12   12
              Prof. Younghee Lee
Bit-round Fair Queuing(BRFQ)

                                  13   13
             Prof. Younghee Lee
      Generalized Processor Sharing
   Motivation
    – BRFQ can’t provide different amount of the capacity to different flows.
    – Differential allocation capacity;
        » To support QoS transport
   GPS
    – provides a means of responding to different service requests
    – each flow  is assigned a weight  that is the number of bits transmitted
      from the queue during each round.
                        Pi
          Fi  Si              ;
          Si  max[Fi1 , R( i )]

          gi          C ; service rate

          Di       where Di  max delay,    Ri  token rate  i ,   Bi  Bucket size
                                                                                         14   14
                                           Prof. Younghee Lee
        Weighted Fair Queuing(WFQ)
     WFQ
      – provide different amount of the capacity to different flows.
      – designed to emulate GPS
      – rule: whenever a packet finishes transmission, the next packet sent is the
        one with the smallest value of Fi.
           Pi
Fi  Si           ;
Si  max[Fi1 , R ( i )]

    B ( K i  1) Li      Ki
                            Lm ax
Di  i 
    Ri      Ri
                         1 C
                         m   m

where Di  max delay,         Ri  token rate  i ,   Bi  Bucket size
       K i  number of node,        Li  max packet _ size,    Cm  outgoing link capacity at node m
       Lm ax  max packet _ length for all flows through all nodes on the path of flow i

                                                                                         15     15
                                               Prof. Younghee Lee

 Isolation
   – Isolates well-behaved from misbehaving sources
 Sharing
   – Mixing of different sources in a way beneficial to all

 Mechanisms:
   – WFQ
      » Great isolation but no sharing
      » fairness
   – FIFO
      » Great sharing but no isolation
      » Efficiency

                                                         16   16
                               Prof. Younghee Lee
      Resource Reservation (RSVP)
   RSVP Characteristics(1)
    – Unicast and multicast: make reservation for both
    – simplex:
    – receiver-initiated reservation
        » ATM, FR: the source requests a given set of resources
             reasonable in a unicast environment but inadequate for multicasting;
              different resource requirements(subflow)
             QoS requirements of different receivers may differ depend on the output
              device, processing power, and link speed of the receiver.
        » A sender provides the routers with the traffic characteristics. The receivers
          specify the desired QoS.
    – maintaining soft state in the internet
        » hard-state approach: a connection-oriented scheme:, fixed route
        » soft-state approach: RSVP
             reservation state

                                                                             17    17
                                     Prof. Younghee Lee
       Resource Reservation (RSVP)
   The call setup process and per-element call behavior

                                                           18   18
                             Prof. Younghee Lee
       Differentiated Services
   Standardizing “Services” or Packet Forwarding “Behavior”?
    – To deploy a new service, you have to upgrade the world
    – A router can’t actually do many different things to a packet
    => standardizing forwarding behavior(“send this packet first” or “drop this
       packet last”)
    => Behaviors + Rules = Services => flexibility
    => No per flow state, resource reservation => scalability
    => Better-than-best-effort service to applications, without the need for
       host RSVP signaling(Few hosts in today’s Internet are able to generate
       RSVP signaling. Users may only want to specify a more qualitative
       notion of the service they require)
    * Think in terms of IP Forwarding/Routing architectural separation

                                                                       19    19
                                   Prof. Younghee Lee
      Differentiated Services
   Edge functions: packet classification and traffic conditioning
    – packet marking, forwarding immediately/delayed/dropped
    – passes: VIP pass...
   Core function: forwarding according to per-hop behavior
    – packet marking: the class of traffic(the behavior aggregate)
    – obviates the need to keep router state for individual source-
      destination pairs

                                                                      20   20
                                   Prof. Younghee Lee
      Differentiated Services
   Traffic Classification and conditioning
    – DS field: IPv4 Type of Service field, IPv6 Traffic Class field
    – DSCP: per-hop behavior; class of traffic
    – simple packet classification and marking

    – packet classification and traffic conditioning at the edge

                                                                       21   21
                                   Prof. Younghee Lee
         Differentiated Services
   Edge Router
    – Traffic Conditioning Agreement(TCA): describes the rules and traffic
      profiles required for conditioning
    – Classification: Identifying the flow the packet belongs to
    – Metering: Measuring the temporal properties of a flow
    – Shaping: Delaying and/or dropping packets so that a flow confirms to
      it’s profile.
    – Marking: Setting the code point in the packets that have been shaped
    – Edge router need to understand RSVP
       » interoperability with IntServ domains must be ensured
       » RSVP can be used as a signaling mechanism for provisioning and configuration.
    – Service Level Agreement(SLA): an agreement between a customer
      and the DS domain. TCA is a subset of SLA
       » other details: Routing constraint, encryption requirements etc.
       » Edge routers are responsible for interfacing with the customers by executing SLAs
       » Bandwidth Broker helps the Edge routers in the admission control of the SLAs.
                                                                                  22         22
                                       Prof. Younghee Lee
        Differentiated Services
   Per-Hop behavior
    – a description of the externally observable forwarding behavior of a DS node
      applied to a particular DS behavior aggregate
        » PHB defines differences in performance
        » does not mandate any particular mechanism for achieving these behavior
        » differences in performance must be observable, and measurable
    – Expedited Forwarding: can be used to construct services with quantitative
        » EF PHB specifies that the departure rate of an aggregate class of traffic from a
          router must equal or exceed a configured rate
        » minimum guaranteed link bandwidth
    – Assured Forwarding
        » four classes: each class is guaranteed to be provided with some minimum amount of
          bandwidth and buffering
        » within each class, packets are further partitioned into one of three “drop preference”
        » Could be used as a building block to provide different levels of service to the end
          systems: Olympic-like service; qualitative services like better than best effort

                                                                                      23     23
                                         Prof. Younghee Lee
       Differentiated Services
   Core Router
    – DS Codepoint(DSCP): Edge routers classify and stamp the packets
      with appropriate DSCP
    – DSCP is translated into a PHB
       » can be many codepoint to one PHB
       » may vary from one DS domain to another
    – PHB:
       » Basic building block for service construction.
       » Resources are allocated to PHB

                                                                 24     24
                                     Prof. Younghee Lee
       Differentiated Services
   Bandwidth Broker
     – Responsible for resource management in a DS-domain.
     – Repository for domain wide policies.
     – Serves as an authenticating agent for users.
     – Maintains the global state of the DS domain.
     – Make sure that the number of simultaneous uses of the PHBs fit within
       the resource allocation. Also helps the edge routers in admission control
     – controls provisioning and configuration of all nodes.
     – Actual implementation can be
         » Centralized: easy to implement; but known problems in terms of performance,
           scalability, fault tolerance
         » Distributed: hard to implement and get it right
     – Can be thought as the brain or the control center for a DS domain.

                                                                                   25    25
                                        Prof. Younghee Lee
       Differentiated Services
   Service Creation
    – DS architecture, lets you create a wide variety of services.
   Service
    – The overall treatment of a customer’s traffic within a DS domain or from
    – can be
        » Quantitative: virtual leased line
        » Qualitative: Olympic service(Gold, Silver, Bronze)
        » Neither: BW allocated to class A is always double that of class B

   Scope of service: The topological extent of the service
    – From a given ingress point
        » to a given egress point, to a given set of egress point, to any egress point.(Open-
          ended Scope)
    – From a egress point
        » from a given ingress point, from a given set of ingress point, from any ingress
          point(open-ended point)                                                      26       26
                                         Prof. Younghee Lee
        Differentiated Services
   Research Issues
     – BB: Centralized Vs. Distributed.
         » Efficient collection and maintenance of global state of the domain
         » coordination and consistency in case of distributed implementation
         » suitable algorithms for completely automated resource management and
           admission control.(measurement based?)
     – SLA and TCA
         »   details that go into SLA/TCA and their format
         »   protocols for automatic SLA negotiation
         »   accommodating dynamically changing SLAs
         »   SLA conflict resolution
     – Implementation of PHBs(E.g.: EF)
         » choice of scheduler
         » buffer management strategies(how much to allocate?)
     – multicast: dynamic multicast groups make provisioning difficult
     – security: Denial of service attacks are easy                      27       27
                                        Prof. Younghee Lee
  Differentiated Services
 RED with In or Out (RIO)
  – Has two classes, “In” and “Out” (of profile)
     » “Out” class has lower Minthresh,
          packets are dropped from this class first

          Based on queue length of all packets

  – As avg queue length increases, “in” packets
    are also dropped
     » Based on queue length of only “in” packets

                                                       28   28
                               Prof. Younghee Lee
      Approaches for QoS in the Internet
   IPv4 TOS: not widely implemented in the current systems
     – service request is associated with individual packet rather than sequence
        of packets. So may not be meaningful always.
     – Service offerings have been tied to the implementation
   Diffserv: Layer 3 only
   Label switching: MPLS: specifies ways that layer 3 traffic can be mapped to
    CO layer 2 transports like ATM and FR
     – QoS state is set up on a Hop-by-Hop basis
     – also helps in speeding up of forwarding process
     – overhead of setting up and maintaining the labeled paths
   Intserv/RSVP
   Cost:
   Compatibility:
     • MPLS can be an intra-domain implementation technology
                                                                      29    29
                                   Prof. Younghee Lee
   Packets sent by a sender are received by more than one receiver.
     – Network replicates the packet
     – Limits the communication overhead on the sender, making it possible to
       send to a large number of receivers
     – Potentially reduces bandwidth consumption in the network                     R
   Union” of point-to-point paths.                                             R
     – combine message over shared links                                            R
                                                   S            R
   Many issues/challenges.                                            R        R
     – How (receiver) multicast group membership managed?
     – How do we route packets?                                                     R

     – How do routers forward multicast packets?
 “Optimizations possible but difficult if the receivers are not known.
     – model used in internet

                                                                           30       30
                                   Prof. Younghee Lee
   The method in the multicast strategy
    – least-cost path from source to each network that includes member of
      the multicast group: spanning tree with networks containing group
    – The source transmit a single packet along the spanning tree.
    – Routers at branch points replicate the packet.

                                                                    31      31
                                 Prof. Younghee Lee
      Requirements for Multicasting
   A convention for identifying a multicast address
    – IPv4: class D addresses. 32-bit; 1110 + 28-bit group identifier
    – IPv6: 128-bit; 8-bit prefix(all 1s)+4-bit flags+4-bit scope+112-bit group id
        » flags: permanently assigned or not
        » scope field: ranging from a single subnetwork to global
   A router must translate between an IP multicast address and a
    subnetwork multicast address
   Individual host informs routers of its inclusion in and exclusion
    from the group for dynamic multicast address generation.
   Routers must exchange two sorts of information: => routing
    – Subnetworks include members of a given multicast group.
    – Information to calculate the shortest path
   A routing algorithm to calculate shortest paths
   Each router must determine multicast routing paths on the
    basis of both source and destination addresses
   Anonymity
   Dynamic join/leave         Prof. Younghee Lee
                                                           32                  32
         Multicast Routing
   Source-Based Tree
     – Shortest path tree for each sender: DVMRP, MOSPF, PIM-DM
     – Shortest path trees – low delay, better load distribution
     – More state at routers (per-source state)
     – Efficient for in dense-area multicast
   Group-shared tree
     – center-based approach: center node( rendezvous point or core)
         * How to select the center?
     – Steiner Tree problem: finding a minimum cost tree: not popular
         * information needed: all links in the network, must rerun whenever link costs change,
          performance is but one of many concerns
     –   Higher delay (bounded by factor of 2), traffic concentration
     –   Choice of core affects efficiency
     –   Per-group state at routers
     –   Efficient for sparse-area multicast
   Major concern might be: extra state in routers
                                                                                         33       33
                                             Prof. Younghee Lee
   Group Management
 Management         strategy depends on usage.
   – how quickly does membership change?
   – restrictions on membership
   – size of the group
 Internet:focus on large groups that can change
  rapidly and with little control over membership.
   – distributed algorithm (scalability, again)
   – receiver-initiated management (scalability, again)
      » Receiver initiated reliable multicast protocols(end-to-end); NACK
   – sender does not have list of receivers (scalability, again)
                                                                        34   34
                                 Prof. Younghee Lee
        Group Management
   Internet Group Management Protocol (IGMP).
    –   Routers jointly keep track of membership
    –   Relies on multicast (e.g. Ethernet) in leaf networks
    –   Protocol defines how receivers contact routers to join a group
    –   Operates locally between a host and an attached router
    –   IGMPv2 Message types
         » Membership query
         » Membership report
         » Leave group
    – Feedback suppression
         » After receiving a membership_query message and before sending a
           membership_report message, a host waits a random amount of time between 0 and
           the maximum response time value defined in IGMP message.
         » Some other attached host reports -> suppress(discard) its own pending
    – Soft state
    – Joining a multicast group: receiver driven

                                                                               35     35
                                       Prof. Younghee Lee
      Routing Approaches
   Create a spanning tree to all routers and prune the tree for each specific
    multicast group
     – Begin by flooding traffic to entire network
     – pruning critical to reduce traffic. (Grafting)
     – scaling is a concern
     – Examples: DVMRP, PIM-DM
     – Unwanted state where there are no receivers
   Link-state multicast protocols
     –   Routers advertise groups for which they have receivers to entire network
     –   Compute trees on demand
     –   Example: MOSPF
     –   Unwanted state where there are no senders
   Core-based multicast routing
     –   create tree for each multicast address with root in the center of the network
     –   multicast messages are sent to the root, which forwards them down the tree
     –   scales better, but potentially less efficient and less robust
     –   CBT, PIM-SM

                                                                                         36   36
                                           Prof. Younghee Lee
    Routing Protocol
   Routing protocols
    –   IGMP: a protocol that enable hosts to join and leave multicast group
    –   DVMRP: Distance Vector Multicast Routing Protocol.
    –   MOSPF: extension to the OSPF for multicast routing within an AS.
    –   PIM: Protocol Independent Multicast
    –   BGMP: for interdomain multicast routing.

                                                                     37    37
                                 Prof. Younghee Lee
        Routing: Group shared tree
   Single routing tree for the entire multicast session
   Steiner Tree problem: finding a minimum cost tree: not popular
    * information is needed about all links in the network
    * needs to be re-run whenever link costs change
   Center-based approach: center node(rendezvous point or core)
    * process used to select the center
         - chosen so that the resulting tree is within a constant factor of optimum
    * CBT, sparse-mode PIM, BGMP
A single,       two source-based   A minimum cost             Constructing a
shared tree     tree               multicast tree             center-based

                                                                          38   38
                                    Prof. Younghee Lee
        Core-Based Trees(CBT)
   CBT multicast routing protocol
    – group-shared tree with single core
       » Unidirectional tree/ bi-directional tree, Core placement/selection,
         Multiple core, Dynamic core…
    – Core forwards over multicast tree
    – Operation
       » sends a JOIN_REQUEST message towards the tree core
       » The core(or the first router that receives the message) respond with
       » maintained by having a downstream router send keepalive
       » immediate upstream router responds with ECHO_REPLY message
       » FLUSH_TREE: if no ECHO_REPLY received

                                                                     39    39
                                  Prof. Younghee Lee
       Routing: Source-based tree
   Shortest path tree: DVMRP, Dijkstra’s algorithm
    – requires that each router know the state of each link in the network
    – compute the least cost path tree from the each source to all
    – Good delay property, Per source and group overhead

                                                                     40      40
                                 Prof. Younghee Lee
        Routing: Source-based tree
   RPF(reverse path forwarding)
    – When a router receives a multicast packet with a given source address, it
      transmits the packet on all of its outgoing links(except the one on which it
      was received) only if the packet arrived on the link that is on its own
      shortest path back to the sender.
    – Otherwise the router simply drops the incoming packet without
      forwarding it on any of its outgoing links. => avoid flooding loop
        » Need to know unicast shortest path to the sender. Not the shortest
          path from the source to itself(assumption:symmetric). Asymmetric

                                                                        41    41
                                   Prof. Younghee Lee
      Routing: Source-based tree
   RPF(reverse path forwarding)
    – RPB: Reverse Path Broadcasting
    – TRPB: Truncated Reverse Path Broadcasting: router truncate its
      transmission to the local network if none of the hosts attached to the
      network belong to the multicast group. Leaf router only
    – RPM: Reverse Path Multicasting: with IGMP
        » pruning: A multicast router that receives multicast packets and has no
          attached hosts joined to that group will send a prune message to its
          upstream router.
              If there were 1000 routers downstream from D; (initial Mbone)

              Grafting message to its upstream router to cancel its earlier prune

                                                                              42     42
                                      Prof. Younghee Lee
Internet Group Management Protocol (IGMP)
   IGMP: used by hosts and routers to exchange multicast group membership
    information over a LAN
   Message format: Figure 15.4
     – version, type, checksum, group address(0 in a request message, valid
        group address in a report message)
   IGMP Operation
     – to join a group: host sends an IGMP report message.
        » Group address field: destination address field of IP header
        » All member hosts will receive the message, and learn of the new member.
    – to maintain a valid current list: multicast router periodically issues a
      IGMP query message, sent in an IP datagram with an all-hosts multicast
      address. Must respond with a report message to remain a member.
    – Multicast router needs to know that there is at least one group member
      still active. Not need to know the identity of every host in group.
        » Any host hears: if some host reports -> cancels report. if no report within the
          timeout -> sends report.
    – Group Membership with IPv6
        » IGMP:IPv4
        » ICMPv6: includes all of the functionality of ICMPv4 and IGMP.
          * ICMP: Internet Control Message Protocol
                                                                              43     43
                                      Prof. Younghee Lee
      Distance Vector Multicast Routing Protocol
   DVMRP:
    – source-based trees with reverse path forwarding, pruning, and
      grafting. Use distance vector algorithm to compute shortest path
      back to source
        » Not from source to the members
    – Data stream reaches all LANs (possibly multiple times). If a router is
      attached to a set of LANs that do not want to receive a particular
      multicast group, the router can send a "prune" message back up the
      distribution tree to stop subsequent packets from traveling where there
      are no members.
    – Since new hosts may want to join the multicast group at any time,
      DVMRP must periodically re-flood. This creates a scaling problem,
      especially if pruning not effective or not implemented.

                                                                      44    44
                                   Prof. Younghee Lee
       Distance Vector Multicast Routing Protocol
   DVMRP implements its own unicast routing protocol (similar to
    RIP) to determine which interface leads back to the source of
    the data stream. The path that the multicast traffic follows may
    not be the same as the path that the unicast traffic follows.
    (asymmetric case?)
   DVMRP has been used to build the MBONE by building tunnels
    between DVMRP-capable machines.
   DVMRP: de-facto Interdomain multicast protocol
   DVMRP is state of the art today.?

                                                            45   45
                              Prof. Younghee Lee
        Multicast Extensions to Open Shortest Path
        First (MOSPF)
   MOSPF: enhancement to OSPF for the routing of IP multicast
    datagram within an AS.
   MOSPF works only in internetworks that are using OSPF.
    – MOSPF is best suited for environments that have relatively few
      source/group pairs active at any given time. It will work less well in
      environments that have many active sources or environments that have
      unstable links.
   Operation:
    – Each router floods information about local group membership to all other
      routers in its area.(Each router attached to a LAN uses IGMP to maintain
      a correct picture of local group membership).
        » Using Dijkstra’s algorithm, each router constructs the shortest-path spanning
          tree from a source network to all network containing members of a multicast
          group.; done only on demand. (When it receives a multicast datagram)
    – For any hop that is across a broadcast network such as LAN, an IP
      multicast datagram is transmitted inside a MAC-level multicast frame.
   Equal-cost Multipath Ambiguity: tiebreaker rule
                                                                             46    46
                                     Prof. Younghee Lee
       Multicast Extensions to Open Shortest Path
       First (MOSPF)
   Interarea multicasting
     – OSPF: backbone, area, border router
     – Each router within a area only knows about the multicast groups that have
       members in its area.
     – Interarea multicast forwarder:
         » subset of an area’s border routers
         » forward group membership information and multicast datagrams between areas
               receives the multicast link status reports, knows all of the multicast group in the
               backbone routers exchange the information on multicast group
               also wild-card multicast receiver. Receive all multicast datagrams generated in
                an area

                                                                                       47      47
                                          Prof. Younghee Lee
       Multicast Extensions to Open Shortest Path
       First (MOSPF)
   Inter-AS multicasting
     – MOSPF has no responsibility for multicasting beyond its AS.
     – Responsible for providing multicast group information to outside entities and for
       accepting multicast datagrams for groups contained within its AS.
         » Boundary router: inter-AS multicast forwarders.(+ MOSPF + OSPF)
         » It receive all multicast datagrams from within the AS; wild-card multicast receiver
         » reverse-path routing: to get the knowledge of the source of a datagram.
                Assumes that source X (outside the AS) will enter the MOSPF AS
                use to send a unicast datagram to X

                                                                                        48       48
                                          Prof. Younghee Lee
       Protocol Independent Multicast (PIM)
   PIM
    – to provide a more general solution to multicast routing.
    – Independent of any existing unicast routing protocol
    – designed to extract needed routing information from any unicast routing
      protocol and may work across multiple ASs with a number of different
      unicast routing protocol
    – supports two different types of multipoint traffic distribution patterns.
   PIM strategy
    – many multicast members, many subnetworks within a configuration have
      the members of a given multicast group => frequent exchange of group
      membership information is justified => data-driven
    – widely scattered members => flooding of multicast group information is
      inefficient => receiver-driven => a center based approach
    – dense-mode protocol: for multicast routing within AS; potential alternative
      to MOSPF. uses flood-and-prune Reverse Path Forwarding and looks a
      lot like DVMRP. However, dense-mode PIM is that PIM works with
      whatever unicast protocol is being used
    – sparse-mode protocol: for inter-AS multicasting routing??
                                                                       49    49
                                   Prof. Younghee Lee
      Protocol Independent Multicast (PIM)
   Sparse-Mode PIM
     center-based approach
    1. For a multicast group, a router is designated as a rendezvous point (RP)
    2. A group destination router sends a Join message to the RP. Requesting
       router uses a unicast shortest-path route to send message. The reverse
       of path become part of the distribution tree from RP to destinations.
    3. A group source router sends packets to RP using unicast shortest-path
    – From RP to the multicast receivers, shared tree is used minimizing the
       number of packets replicated.
    – PIM allows a destination router to replace the group-shared tree with a
       shortest-path tree to any source.(source-specific tree); Once it receives a
       packet from the source, it send a Join message back to the source router
       => sends Prune message to RP
    – The selection of an RP is dynamic.
    – RP placement is not a critical issue.

                                                                        50    50
                                   Prof. Younghee Lee
Protocol Independent Multicast (PIM)

                                    51   51
               Prof. Younghee Lee
       Inter-domain Multicast Routing(BGMP)
   The case that different AS’s choose to run different multicast routing
     – IETF idmr working group
     – DVMRP: defacto interdomain multicast routing protocol
         » not well suited to the sparse set of routers participating in today’s Internet
     – group-shared tree approach toward routing
         » problem: a center could conceivably be chosen in a domain that does not
           contain any hosts in the multicast group : third party dependency(No
           problems in the intradomain case)
              “unfairly” burden the domain which has no interest in the multicast group

              performance dependencies on domains outside of those participating in
               the group

                                                                                  52        52
                                        Prof. Younghee Lee
        Overlay Multicast: ALM
 Potential     benefit over IP multicast
   – Quick deployment
   – All multicast state in end systems
   – Computation at forwarding points simplifies
     support for higher level functionality
 Concerns
  – closely matched to real network topology to be efficient?
  – Performance
      » Increase in delay
   – Bandwidth waste (packet duplication)

                                                         53     53
                            Prof. Younghee Lee
    Mobile IP
   Communicate with mobile hosts using their home IP
    – should be transparent to applications and higher level protocols
    – minimize changes to host and router software
   Each area has a home agent and foreign agent that
    managing packet forwarding.
    – binding = (IP address, foreign agent address)
    – binding includes time stamp
   Try to short circuit the home location by going directly to
    the foreign agent.
    – cache bindings in the appropriate places
    – protocol to update/invalidate caches
    – security considerations
                                                                   54    54
                               Prof. Younghee Lee
     Mobile IP in IPv4
   Registration process
     – mobile host registers with home
       and foreign agent
     – cache bindings: address, care-of-
       address                                             Home
   Tunneling is used to forward                           Agent
    packets between agents.
   Supporting mobility
     – invalidating old caches explicitly or                            Source
       in a lazy fashion                            Foreign
                                                    Agent 1
   Many variants and optimizations
     – Mobile host can be its own foreign
       agent, e.g. can get local addresses
     – Source can redirect packet directly                    Foreign
       to foreign agent                                       Agent 2

                                                                           55    55
                                      Prof. Younghee Lee
Mobile IP Goals
   IP address encodes the host’s network.
    – Simplifies routing in the common case: look only at network
      identifier, but not not at the host id
    – Makes special cases hard, e.g. what happens when the host
   Communicate with mobile hosts using their “home” IP
    – should be transparent to applications and higher level protocols
   Minimize changes to host and router software
    – No changes to communicating host
   Security should not get worse.

                                                                56       56
                             Prof. Younghee Lee
    Mobile IP (IPv4)
   Home network has a home agent that is responsible for
    intercepting packets and forwarding them to the mobile
    – E.g. router
    – Forwarding is done using tunneling
   Remote network has a foreign agent that manages
    communication with mobile host.
    – Point of contact for the mobile host
   Binding ties IP address of mobile host to a “care of”
    – binding = (IP address, foreign agent address)
    – binding includes time stamp

                                                            57   57
                                Prof. Younghee Lee
     Mobile IP Operation
   Agents advertise their presence.
     – Using ICMP or mobile IP control
     – Mobile host can solicit agent
     – Mobile host can determine where it
   Registration process: mobile host
    registers with home and foreign                                     Source
    agent.                                          Foreign
                                                    Agent 1
     – Set up binding
   Tunneling
     – forward packets to foreign agent
     – foreign agent forwards packets to
       mobile host                                            Foreign
                                                              Agent 2
   Supporting mobility
     – invalidating old caches in a lazy
       fashion                                                          58   58
                                     Prof. Younghee Lee
 Mobile   host can be its own the foreign agent.
  – mobile host acquires local IP address
  – performs tasks of the mobile agent
 Short  circuit the home location by going directly
  to the foreign agent.
  – Routers in the network store cache bindings and
    intercept and tunnel packets before they the mobile
    host’s home network
  – Need a protocol to update/invalidate caches
  – Raises many security questions and is not in the
                                                    59    59
                      Prof. Younghee Lee