Term
|
Definition
- Routes datagrams to a single member of a group of potential receivers that are all identified by the same destination address
- An anycast address can be used to find the nearest instance of a service
- Useful for discovering the nearest instance of a service or load-balancing distributed services
- Many hosts can be configured to accept IP traffic to the same anycast address
- Such hosts can also have a normal IP address
- Routing system carries a route for the address
- Packet only received and processed by one system
|
|
|
Term
| What are the principles for IP multicast packets |
|
Definition
- Look like normal IP packets, however:
- IP destination represents a group of hosts
- Uses the range 224.0.0.0 - 239.255.255.255
- Group can (dynamically) contain any # of hosts
- Group members can be anywhere on network
- They are forwarded by routers towards known receivers
- Same multicast packet is never sent twice on the same link, hence the bandwidth used on a specific link is independent of the number of receivers
- Routers keep multicast forwarding state to know which interfaces they forward packets on for a given group address
- A set of data paths for packet in a multicast routed network is often referred to as a multicasst tree
- Routers are branches and receivers are leaves
|
|
|
Term
| What are some scenarios where multicast would be useful? |
|
Definition
- Campus or enterprise Internet Protocol TV (IPTV)
- Typical consumer-oriented use of multicast
- An ADSL user could send live video to 1000s of users
- Number of receivers not an issue
- Multi-party applications (Conference/gaming)
- Where same data is sent to multiple participants
- For financial applications
- Deliver information quickly and simultaneously
- Software updates
- Also used transparently for (local) service discovery
|
|
|
Term
| Characteristics of sending IP multicast |
|
Definition
- IP Multicast traffic looks like regular IP (UDP) traffic
- Destination address is not a single fixed hsot
- IP represents a set of potential receiving host who express interest in receiving traffic sent to that group address
- Receiving hosts might change over time
|
|
|
Term
| Characteristics of IPv4 Multicast Addressing |
|
Definition
- Class D = 224.0.0.0 - 239.255.255.255
- Subdivided in rather complex ways
|
|
|
Term
| How to pick a group address multicast to use? |
|
Definition
- 239.255.0.0/16 for local use within a site
- Implies traffic to these destinations is filtered at your site border
- For sending multicasts externaly
- Need unique global scope multicast addresses (just like you need globally unique addresses for unicast)
- But no hierarchical address allocations for multicast like there is for unicast
- Two options you can use:
- GLOP
- Unicast prefix-based addresses
|
|
|
Term
| What are the options to send external multicasts (outside your site)? |
|
Definition
- GLOP: 233.x.y.0/24, where x.y is your 16-bit BGP AS number
- Unicast Prefix-Based addresses: 234.a.b.0/24, where a.b.0.0/16 is your unicast allocation
|
|
|
Term
| How do receivers know which group to join? |
|
Definition
- In order for an application to receive multicast traffic, it needs to find out which group address to use
- Might be built to the application (transparent)
- Information may be available via internet
- Specific 'listing' protocol can be used, Like SAP
- Session Announcement Protocol (SAP)
- Allows senders to announce multicast content using multicast
- Announcements sent to well-known global group, 224.2.127.254
- A SAP announcement doesn't necessarily mean the content is there
|
|
|
Term
| What is Session Announcement Protocol? |
|
Definition
- Protocol for broadcasting multicast session information
- Allows senders to announce multicast content using multicast
- Announcement sent to a well-known global group
- Announcement doesn't necessarily mean the content is there
|
|
|
Term
| What are the principles of the multicast service model? |
|
Definition
- Anyone can send to the multicast group
- Sender(s) don't need to know where receivers are or how many
- They just send packets to the group addresses
- Hosts interested in receiving content for the group join it
- Hosts only signal interest in receiving a group to their default router
|
|
|
Term
| What challenge does routing impose on multicasting? |
|
Definition
- If multicast hosts can be anywhere (only telling their local default routers which group they are interested in) how can other routers on the network learn form where they should accept multicast and to where they should forward multicast?
- Routers need to learn and build multicast forwarding state
|
|
|
Term
| What are some characteristics of Unicast Routing? |
|
Definition
- Router runs a routing protocol
- Link-state with a site, e.g. OSPF
- Routing tables are built
- includes destination prefix, interface, cost
- Router can then forward IP packet appropriately
- Next router then makes forwarding decision
|
|
|
Term
| What are some characteristics of Multicast Routing? |
|
Definition
- Multicast router sees destination of group address G
- Consults multicast forwarding tables to see which interfaces to send multicast packets to group G
- Table includes destination group, interfaces
- Multicast routers may replicate packets
- My default, multicast is not forwarded by a router
- Need for a method to build forwarding table
|
|
|
Term
| What are some solutions to carry out Multicast Routing? |
|
Definition
- Flooding multicast over an entire site network
- Ensures everyone gets it (EVERYONE)
- Process adds multicast forwarding state
- Pruning is a way of restricting the flooding to where forwarding is wanted
- Subnet routers need to know which groups their locally connected hosts want to receive
- They know when to keep forwarding and when they can prune
- This is done using internet group management protocol (IGMP) between hosts and routers
|
|
|
Term
| What problem comes up with flooding, and how to solve it? |
|
Definition
- Can cause infinite loops
- Reverse Path Forwarding (RPF) check can be used
|
|
|
Term
| What is Reverse Path Forwarding (RPF) check? |
|
Definition
| A router discards a multicast packet if it isn't received in the interface which it would use for sending unicast packet to the original sender of the packet |
|
|
Term
| Give an example of Reverse Path Forwarding (RPF) check |
|
Definition
|
|
Term
| What are the main characteristics of the Internet Group Management Protocol (IGMP)? |
|
Definition
- A host sends an IGMP report when it wants to join a group
- This is multicast on the local subnet; will be seen by the router
- This informs default router of host's interest in a group
- A host also sends an IGMP message when it leaves the group
- Router may then no longer need to forward multicast on that subnet
- It checks for other listeners before it stops forwarding multicast
- Router sends periodic IGMP queries on the link (Default 125 sec intervals)
- Hosts respond with which groups they want to receive
- Groups no longer forwarding by router if no responses to the queries
- This catches hosts that crashed and never send a leave message
|
|
|
Term
| What are characteristics of Multicast and Ethernet? |
|
Definition
- IPv4 Multicast destinations are mapped to Ethernet addresses with prefix 00:01:5e:00:00:00 through 00:01:5e:7f:ff:ff
- Lower 23 bits of multicast group are mapped to MAC address
- Switches by default forward multicasts on all ports
- Treat multicast frames like broadcast frames
- All hosts on same subnet will see same multicast traffic in switched network
|
|
|
Term
| How can we limit which hosts at Layer 2 receive multicast packets? |
|
Definition
- Avoid extra load on non-interested hosts, so they don't need to process the multicast packets
- Desirable to minimise this particularly for higher-bandwidth multicast streaming where different hosts in a subnet may be interested in different streams
- Important where lot of high bandwidth channels exist
- IGMP snooping offers a solution
- Switches observe IGMP messages between hosts and routers and only multicast to hosts that want it
|
|
|
Term
| What is does IGMP Snooping do? |
|
Definition
| Limits which hosts at layer 2 receive multicast packets by having switches observe IGMP messages between hosts and routers and only forward multicast to hosts that want it |
|
|
Term
|
Definition
- Routers that have no interested downstream do not need to receive multicast, so can inform their upstream router to no longer multicast to them
|
|
|
Term
| What does Repeated Flood-and-Prune consist of? |
|
Definition
- Initially packets for a group are flooded everywhere
- Routers at edge can prune back
- Leads to creation of "multicast tree" for group
- Routers remember prune state for period of time. When it expires, the flooding will be repeated downstream (And pruned if necessary)
- Allows for topology changing (broken nodes, etc)
- If new host joins the group, their default router can send a unicast 'graft' message upstream
- Knows the interface to use (RPF) from previous flooding
- Avoids host having to wait for re-flooding
|
|
|
Term
| When is Flood-and-Prune good? When is it bad? |
|
Definition
- Ok if receivers are denseley populated
- Not so good for a sparse population of receivers (e.g. the internet)
|
|
|
Term
| What is and what are the types of Protocol Independent Multicast (PIM)? |
|
Definition
- Protocol which uses unicast routing table to do RPF, where the protocol chosen does not matter
- Dense Mode - Pretty much flood-and-prune with minor improvements. Usually only used within local site networks
- Sparse Mode - Commonly used protocol, more suitable to internet use
|
|
|
Term
| What are some basic principles of the Protocol Independant Multicast - Sparse Mode (PIM-SM) ? |
|
Definition
- Designed to work w a sparse population of receivers
- Receivers only know group address, not sender's IP
- Makes use of Rendezvous Point (RP) between senders and receivers
- All routers in a site configured to use same router as the RP
- Sender's router initially forwards multicast to RP
- Receiver's router sends JOIN messages to RP
- RP thus knows about senders and receivers
- Multicast trees, initially, pass through RP
- Tree is optimised once receiver gets some multicast traffic and sees the sender's IP address - receiver's router can then JOIN directly towards the sender
|
|
|
