Internet Technology. 13r. Pre-Exam 3 Review Paul Krzyzanowski. Rutgers University. Spring 2013

Transcription

1 Internet Technology 13r. Pre-Exam 3 Review Paul Krzyzanowski Rutgers University Spring

2 This is not a complete review just the key concepts that you should know for the exam It also does not cover Monday s lecture on wireless & QoS 2

3 Routing 3

4 Routing algorithms Goal Find the least-cost path from one node to another node Cost = sum of edge costs Two approaches Global routing algorithm Has global state knows the complete graph These are link-state (LS) algorithms Dijkstra s algorithm is an example of these Decentralized routing algorithms Each node knows its own connectivity and exchanges information with neighbors to compute a least-cost path to each destination A distance-vector (DV) algorithm is an example of this u v x w y z 4

5 Link-State (LS): Dijkstra s algorithm Iterative algorithm that finds the lowest cost to each node from one specific node. You must know the entire network graph Keep track of: A list of nodes for which we already found the least cost (initially just the current node) For each node v in the graph: D(v): Distance our current knowledge of the least cost to get to node v p(v): Previous The previous node before v along that least-cost path to node v Initially, we look at the neighbor of our starting node u D(v) is set to the cost of the edge from our starting node u to v p(v) is set to u (the path is just one edge) All non-neighbor D(v) values are set to infinity After k iterations, least-cost paths are known to k nodes 5

6 Dijkstra s algorithm Initially, we look at the neighbor of our starting node u D(v) is set to the cost of the edge from our starting node u to v p(v) is set to u (the path is just one edge) All non-neighbor D(v) values are set to infinity Each iteration 1. Find a node n that has the smallest distance, D(n), that we have not yet put into our least-cost list 2. Add the node to the least-cost list 3. for each neighbor m of that node The cost of going through node n to m is the cost to n + the cost of the edge nm D(n) + c(n, m) If this cost is smaller than the value we currently have for D(m) Update the cost to the node, D(m), and set p(m) = n After k iterations, least-cost paths are known to k nodes 6

7 Dijkstra s algorithm Since we know the previous node along the lowest cost path, we can compute the route by traversing the path backwards For example, suppose we have these results of least-costs from node a If we need to find the path from a to f, we can work backwards from f Previous node to f = 1 Previous node to d = c Previous node to c = a D(b),p(b) D(c), p(c) D(d), p(d) D(e), p(e) D(f), p(f) 2, a 1, a 2, c 4, d 4, d Least cost path from a to f = a c d f To create a routing table, we will need to know the first hop, so we traverse the path for each destination node and record the first hop First hop at a on the path to f = c 7

8 Oscillations If edge weights represent load on a network link Routers will redirect traffic to lower loads (lower cost) But that will cause those load values to go up! Recalculating minimal cost routes will now favor routes that used to be high cost because they are not loaded anymore The same thing happens now traffic will go over low-load paths, which will make them high-load paths Best approach to avoiding this oscillation is to have routers run the LS algorithm at random times, not synchronized 8

9 Distance-Vector algorithm Distributed algorithm A node (router) only knows about its neighbors A distance vector (DV) is a list of costs to a each node from a specific node Initially, this is the direct cost to a neighbor Non-neighboring nodes are infinity (or not yet in the vector) Each node maintains a distance vector table: DVs from each neighbor A list of costs to get to other nodes in the network Initially, before we receive any DVs, these are all set to infinity 9

10 Distance-Vector algorithm A node n receives a distance vector from a neighbor m It saves the distance vector for m It updates its own distance vector by comparing its values with those in the distance vector it received from m. For each node x: Is the cost of routing through node m lower than the cost I currently have? Cost of routing through m = [ cost(x) from the x s DV ] + [ cost(mx edge) ] If the cost is lower, update its own distance vector with the lower cost If n s distance vector changed, it sends it to its neighbors Keep doing this until the algorithm converges (no changes) Any time the cost of a link changes, a node will send its updated DV to its neighbors and the process restarts 10

11 Poison reverse If a route becomes invalid, a DV algorithm will never converge The updated distance will grow with each iteration Count to infinity! Prevent routing loops If A routes through B to get to C A will always tell B (in its distance vector) that its distance to C is This way B will not try to route through A to get to C if its BC link is down 11

12 Autonomous Systems (AS) Collection of routers and hosts that are administered together (e.g., by one ISP or company) Autonomous systems (AS) connect together to form the Internet They provide a two-level hierarchy machines within an AS and routing between ASes Intra-AS routing protocols = Interior Gateway Protocol (IGP) Internal routing protocols run on these systems. It is up to the administrators to set these up; the rest of the world does not see it or care Protocols such as OSPF and RIP are used within an AS Gateway routers provide links outside an AS to connect to other Inter-AS routing protocol = Exterior Gateway Protocol (EGP) Used allow route to other ASes and between ASes BGP (Border Gateway Protocol) is used here 12

13 Routing Information Protocol (RIP) A form of the Distance Vector algorithm Routing table contains Next router (link) and number of hops to the destination for each destination A node sends RIP advertisements An advertisement is just the routing table Upon receiving the routing table from node N A node chooses to route through N if it will result in fewer hops A node will add any new subnets that it discovers from N 13

14 Open Shortest Path First (OSPF) RIP: based on distance-vector algorithm OSPF: based on link-state algorithm (Dijkstra s shortest path) Each router constructs a complete graph of the systems in the AS Any link changes are broadcast to all routers (not just neighbors) Computes shortest paths What if the AS is huge? OSPF allows a network to be broken into groups of routers Each group is an OSPF Area Routers compute routes only within an area Backbone Area Set of routers that connects areas together To get to other areas, a router will route to a node that is part of the backbone area 14

15 Border Gateway Protocol (BGP) BGP is an Exterior Gateway Protocol (EGP) Used for routing between ASes A pair of gateway routers in each AS maintains a TCP connected These are BGP peers and exchange routing information External BGP (ebgp) session: connection between BGP peers (gateway routers) Internal BGP (ibgp) session: connections within an AS to propagate information on how to route to other ASes (other parts of the Internet) A gateway router sends advertisements of routes Announces what routes it can handle (aggregated as CIDR prefixes) and identifies the path & next hop for that route These advertisements are called prefix reachability If a router learns new prefixes (e.g., from other ASes), it will send advertisements to its connected routers Policies determine what routes an AS may choose to advertise or block BGP is similar to a distance vector protocol but allows policies to specify preferences over costs 15

16 Multicast 16

17 Flooding Uncontrolled flooding Send a packet Each router will duplicate a packet onto each link This can result in the cycles where the same router gets the packet multiple times Sequence number controlled flooding Attach a sequence number to the message Each router keeps track of messages it forwarded If it sees an incoming duplicate, it discards it Reverse path forwarding (RPF) Each router checks the source address on the packet If it came on the link that corresponds to the shortest path to the source, duplicate & forward it; else discard it 17

18 Spanning tree A fully connected graph that contains no cycles Building a spanning tree Pick one node: rendezvous point (or center node) Each node that wants to receive multicasts will send a join message to the rendezvous point As the join message propagates to the node, each link becomes part of the spanning tree. Each router that sees the message will keep track of links that are part of the spanning tree Multicasts will only be sent to those links 18

19 IP multicast IP datagrams have a special destination address A Class D multicast IP address: starts with 1110 Host group = the set of machines that are listening to that address Any machine can send to any class D address IGMP: Internet Group Management Protocol Protocol between a host and its attached router Allows a host to tell a router that it wants to receive data that is sent to a specific multicast address Host sends a membership report message to join Periodically, a router sends a membership query, asking hosts what multicast groups they want to receive: if a host does not respond, the router will delete the membership after a while (this is called soft state) A host may send a leave group message to say it is no longer interested At some point, a router may see that nobody on a link is interested in a multicast address 19

20 IP multicast on the Internet (between routers) IGMP used for routers to talk with hosts on a connected network PIM (Protocol Independent Multicast) used to deliver multicast datagrams across routers PIM Dense Mode Multicast Use a spanning tree with reverse path forwarding (RPF) Flooding is initiated by the sender s router PIM Sparse Mode Multicast A rendezvous point is defined. Interested routers send join messages to that rendezvous point A sender transmits a multicast packet only to the rendezvous point 20

21 Data Link Layer 21

22 Data Link Layer Deals with local area networks and device connectivity MAC = Medium Access Control = protocol for sending/receiving frames at the data link layer Frames are addressed with a MAC address (e.g., ethernet address, not an IP address) MAC protocols may perform error detection and (sometimes) correction Saves extra work for the processor by not sending bad data to the network or transport layers 22

23 Error Detection & Correction Parity Simplest form: 1 bit ensures that total number of bits is odd (odd parity) or even (even parity) Only detects 1 bit errors reliably Two-dimensional parity Simple form of an Error Correcting Code (ECC) Parity bit for each row and each column when data arranged in a grid Can correct a one-bit error (sometimes more) Checksum: treat bits of a packet as integers & perform an operation on them Internet checksum: we saw this in IP, UDP, TCP, ICMP, OSPF, and IGMP headers Weak but simple to compute sum, add 1 for each carry, and complement the result 23

24 Cyclic Redundancy Check (CRC) Know how to compute a CRC (study the text!) Pre-defined generator, G, value r+1 bits long Shift data left by r bits Divide data by the generator using exclusive-or instead of subtraction The remainder (r bits) is the CRC To validate data: Shift data left by r bits and fill in the right with the CRC Divide the data by the generator exclusive-or instead of subtraction The remainder should be zero 24

25 Multiple Access Protocols Multiple Access = multiple nodes need to send messages on the same link Three categories Channel partitioning Time division multiplexing (TDM) Frequency division multiplexing (FDM) Random access Node gets full use of the channel; no scheduled frequency or time slots Chance of collision Taking turns Polling protocol Token passing protocol 25

26 Random Access Protocols Slotted ALOHA Network access is divided into time slots you transmit at the start of one time slot and end at the end of the slot If there is a collision (2 or more nodes transmit in the same slot) Retransmit on the next slot with probability p If you did not retransmit, retransmit on the following slot with probability p Not a very efficient protocol: only 37% of slots end up with good data Carrier Sense Multiple Access with Collision Detection (CSMA/CD) Protocol for Ethernet on a shared channel Listen first; wait for the channel to be clear Then transmit Listen while transmitting if collision, stop transmitting and wait a random time interval Each time a frame experiences a collision pick a random interval from a larger set of wait times: binary exponential backoff 26

27 Ethernet addressing Ethernet addresses are unrelated to IP addresses Address Resolution Protocol (ARP) Finds the Ethernet MAC address for an IP address Broadcast an ARP query with an IP address All adapters receive it If one of the adapters is responsible for the IP address, it sends an ARP response containing the corresponding Ethernet MAC address Keep an ARP table for recently-used IP-MAC mappings IPv6 uses Neighbor Discovery instead of ARP Sends a query (Neighbor Solicitation) to a multicast IP address The multicast IPv6 address will correspond to a multicast MAC address One node will usually get this and respond with its MAC address with a Neighbor Advertisement message Multiple IPv6 multicast addresses may map to one MAC address, so extra nodes may get the message 27

28 Link Layer Multicast Ethernet supports a range of MAC addresses for multicast Both IPv4 & IPv6 have a similar approach Copy 23 bits of an IP class D address to a MAC multicast address Copy 32 bits of a 128-bit IPv6 multicast address to a MAC address An Ethernet card may: listen for a set of addresses pick up all multicast packets pick up all packets In all cases, IP-layer software will have to check that an unwanted multicast datagram did not arrive 28

29 Switched LANs Ethernet switches act like simple routers at the MAC layer Dedicated link to each host (or to another switch) Switch forwards frames from one interface to another Switches are self-learning A switch learns which MAC address corresponds to which interface by looking an MAC source addresses coming in on that interface This information is stored in a switch table Initially, the switch table is empty When a frame is received at an interface, the source address is associated with that interface 29

30 VLANs (Virtual Local Area Networks) Partition a switch to look like several switches Each logical switch is a separate VLAN Broadcasts stay within one VLAN Frames need to be routed between VLANs by a router (often built into the switch) VLAN Trunking Connect multiple VLAN-enabled switches with one cable Extended Ethernet frame has a VLAN tag to identify which VLAN a frame belongs to This tag is stripped before sending it to hosts 30

31 The end 31