Introduction to Communication Networks Spring Unit 15 Internetworking (cont) Routing

Transcription

1 Introduction to Communication Networks Spring 007 Unit 5 Internetworking (cont) Routing EECS SPRING 007

2 Acknowledgements slides coming from: The book by Peterson/Davie The book by Wiliam Stallings Several slides from the earlier issues of the EECS taught by Prof Jean Walrand, A series of slides from lectures by Peter Steenkiste (CMU), also Anja Feldmann(TUBerlin), Nick McKeown (Stanford), and D.Peterson (Princeton) EECS SPRING 007 of 5

3 Remember the picture? Layers and : Ethernet Switch Router Ethernet Switch p Destination Address B Next Hop C Local address C Layer address y Application Transport Destination Address B Local to port p Local address B Layer address w Application Transport Network A Link x Phy Phy Link Phy Phy Network C D Link y Phy Link v Phy Phy Link Phy Phy Network B Link w Phy TOC IP Addressing Examples EECS Layers SPRING / 007 of 5

4 Now: I do have the IP address so what? How to reach another host on the same (sub)network? Consider a IEEE 80. based LAN: Hosts are characterized by their unique MAC address if a host wants to send a packet to another host (or router) within its network it must know the destination host s hardware address. How to find the appropriate MAC address from the IP address? example organization IP address topological MAC address 00:50:56:0:ab: Flat, unique, permanent IP address (ARP) -to- MAC address Answer the ARP service!! EECS SPRING 007 of 5

5 Principles of ARP The ARP exploits the broadcast support. Distributed operation per host. (A special server only if no broadcast!) All hosts can send ARP requests and ARP replies Request: Searched IP, MAC of the sender; Reply: to requestor only. Already resolved addresses are stored in an ARP cache Using soft state principle ) EECS SPRING 007 5of 5

6 IP packet delivery in a subnet [Walrand, ] IP IP on on same subnet IP H e e e IP IP X e H all e e: Who is IP? IP H e IP e R R e5 e e e: I am IP Address Resolution Protocol = Layer Layer Address Layer Layer Address TOC IP Addressing CIDR EECS SPRING 007 6of 5

7 IP Packet delivery not on the same subnet [Walrand, ] Note: IP has explicitly specified thé MAC address e Unlike in bridging the router has to be known to IP (default?) IP H H IP e e e e IP IP X e SH R IP IP X IP IP not on on same subnet e5 e R e5 all e5 I am IP e e5 Note: Fragmentation may be required at R e H IP IP IP X Who is IP? TOC IP Addressing CIDR EECS SPRING 007 7of 5

8 Example: Internet Interconnection How to report errors? AS- AS- AS- Interior router BGP router AS: Autonomous System EECS SPRING 007 8of 5

9 ICMP Internet Control Message Protocol Network-layer above IP: ICMP Used by hosts, routers, gateways to communication network-level information error reporting: unreachable host, network, port, protocol echo request/reply (used by ping) Other functions associated with ICMP are: Reachability testing Congestion Control Route change information Performance measuring Subnet addressing EECS SPRING 007 9of 5

10 ICMP Internet Control Message Protocol ICMP messages are carried in IP datagrams ICMP error message: type, code plus first 8 bytes of IP datagram causing error EECS SPRING of 5

11 ICMP - messages Type ICMP Message Type 0 Echo Reply Destination Unreachable Source Quench 5 Redirect (change a route) 8 Echo Request Time Exceeded for a Datagram Parameter problem on datagram Timestamp Request Timestamp Reply 5 Information request (obsolete) 6 Information reply (obsolete) 7 Address mask Request 8 Address Mask Reply EECS SPRING 007 of 5

12 ICMP destination unreachable Code Value Meaning 0 Network unreachable Host unreachable Protocol unreachable Port unreachable Fragmentation needed and DF set 5 Source route failed 6 Destination network unknown 7 Destination host unknown 8 Source host isolated 9 Communication with destination network administratively prohibited 0 Communication with destination host administratively prohibited Network unreachable for type of service Host unreachable for type of service EECS SPRING 007 of 5

13 An Example Ping Used to detect a hosts reachability by exchanging a requestresponse pair of ICMP packets Type is 8 for request/0 for reply Code is 0 (not used) Identifier allows to differentiate different sessions Sequence number Supports matching the answer to the request. EECS SPRING 007 of 5

14 Traceroute and ICMP Source sends series of UDP segments to dest. (UDP just IP plus something ) First has TTL = Second has TTL =, etc. Unlikely port number When n th datagram arrives to nth router: Router discards datagram And sends to source an ICMP message (type, code 0) Message includes name of router & IP address When ICMP message arrives, source calculates RTT Traceroute does this times Stopping criterion UDP segment eventually arrives at destination host Destination returns ICMP host unreachable packet (type, code ) When source gets this ICMP, stops. EECS SPRING 007 of 5

15 Example: Internet Interconnection How to support mobility? AS- AS- AS- Interior router BGP router AS: Autonomous System EECS SPRING of 5

16 Motivation for Mobile IP Routing based on IP destination address, network prefix (e.g. 9..) determines physical subnet change of physical subnet implies change of IP address to have a topological correct address (standard IP) or needs special entries in the routing tables Specific routes to end-systems? change of all routing table entries to forward packets to the right destination does not scale with the number of mobile hosts and frequent changes in the location, security problems Changing the IP-address? adjust the host IP address depending on the current location almost impossible to find a mobile system, DNS updates take to long time TCP connections break, security problems EECS SPRING of 5

17 Requirements to Mobile IP (RFC 00) Transparency mobile end-systems keep their IP address continuation of communication after interruption of link possible point of connection to the fixed network can be changed Compatibility usage of the same layer protocols as IP no changes to current end-systems and routers required mobile end-systems can communicate with fixed systems Security authentication of all registration messages Efficiency and scalability only little additional messages to the mobile system required (connection typically via a low bandwidth radio link) world-wide support of a large number of mobile systems in the whole Internet EECS SPRING of 5

18 Terminology Mobile Node (MN) system (node) that can change the point of connection to the network without changing its IP address Home Agent (HA) system in the home network of the MN, typically a router registers MN s location, tunnels IP datagrams to the COA Foreign Agent (FA) system in the foreign network of the MN, typically a router forwards the tunneled datagrams to the MN, typically also the default router for the MN Care-of Address (COA) address of the current tunnel end-point for the MN (at FA or MN) actual location of the MN from an IP point of view can be chosen, e.g., via DHCP Correspondent Node (CN) communication partner EECS SPRING of 5

19 How does it work Agent Advertisement HA and FA periodically send advertisement messages into their physical subnets MN listens to these messages and memorizes the HA MN moves MN detects the Advertisement of the FA MN reads a COA from the FA advertisement messages Registration MN registers within the FA (soft state) MN registers his move at the HA, disclosing the COA of the FA HA acknowledges via FA to MN Advertisement of the MN IP address by HA Within the home network HA handles the ARP requests for IP address of MN answering with the own MAC adddress To all others HA advertises the IP Adrress of the MN in the home network. EECS SPRING of 5

20 Data transfer from the mobile system HA MN home network Internet FA sender foreign network CN receiver Sender sends using his usual IP address from the home Network using FA as default router. EECS SPRING of 5

21 Reverse tunneling (RFC ) HA MN home network Internet sender FA foreign network CN receiver. MN sends to FA. FA tunnels packets to HA by encapsulation. HA forwards the packet to the receiver (standard case) EECS SPRING of 5

22 Mobile IP with reverse tunneling Sending packet with his old IP address MN behaves topologically incorrect. Some firewalls might block this Reverse tunneling solves the problem! a packet from the MN with his IP address and the IP Address of the corresponding host is posted to the FA Using it MAC address. FA encapsulates it with its IP Address and destination address HA - Tunnel (IP packet in IP Packet!!!) Side effects: multicast and TTL problems solved (TTL in the home network correct, but MN is to far away from the receiver) Reverse tunneling does not solve Security considerations completely, the reverse tunnel can be abused to circumvent security mechanisms (tunnel hijacking) optimization of data paths, i.e. packets will be forwarded through the tunnel via the HA to a sender (double triangular routing) EECS SPRING 007 of 5

23 Data Transfer to the mobile system COA:Careof Address home network router HA Tunnel router FA MN Internet foreign network CN router home network router HA. router FA.. MN Internet foreign network. CN router EECS SPRING of 5

24 More about tunneling... IMPORTANT Network R Internetwork R Network IP header, Destination =.x IP payload IP header, Destination = IP header, Destination =.x IP header, Destination =.x IP payload IP payload The complete (original header end payload) are now considered new payload!. The new payload can be completly encrypted, and hidden from any kind of observation. Router R can consider Network as directly connected. It might be a subnetwork of location. Virtual private networks use tunnels to connect locations... EECS SPRING 007 of 5

25 ROUTING _ THE THEORY Remember: In general, routing is about finding the way from the source to a given address. In which layer?: While this task is typical for the Network Layer it can also appear in other Layers (remember learning bidges?) The general problem is a graph problem: nodes are components - verges are links Forwarding: Is the processing of the packet in the node, assuming that the routing information is known. EECS SPRING of 5

26 Source Routing vs. Hop by hop routing Source routing: the source node defines the whole route.. Features: No state information is needed in the individual nodes Adapting to link breakdowns causes global consequences... EECS SPRING of 5

27 Hot potato (deflection) routing vs. Targeted routing Targeted routing: This is what we have used so far: there have been tables full of useful information and forwarding used this tables... Reminder for the datagram case... There has always be the port defined for each destination address... Reminder : Using this tables forwarding. Getting these tables prepared Routing Hot potato routing Assume a packet arrives, and there is NO entry for the destination in the tables... Instead of dropping the packet, one might decide to forward it SOMWHERE _ select ANY available port. (Funny the same can be done if the output queue is full!!!) Intuition: If I do not know the way (or cannot accomodate) possibly some other element might know... Just for those curiose ones: Used, e.g. In all optical netowrks with no buffers... EECS SPRING of 5

28 Flooding Packet broadcasting from node A to all other nodes by (a) pure flooding and (b) a spanning tree. Arrows indicate packet transmissions at the times shown (each packet transmission time is assumed to be one unit long). In general, flooding is an approach to achieve quickest possible distribution of information, independent of the topology. It is also nice in the case of unreliable links... In general more packets are send than really needed... EECS SPRING of 5

29 Constrains for flooding Simple flooding implies multiple arrival of a packet... Flooding should not go on infinitely...some limitation is needed Different approaches to constrain the flooding: Lifetime of a packet can be limited (say to N traversed nodes) Nodes could store some parameters of the packet, and deny transmissions which are obviously repetitions. Candidate parameters: Source address, and packet id. Problem: In each case looking into the packet is necessary!!! EECS SPRING of 5

30 Addendum: Broadcast Tree Construction E Flooding is used to construct the tree rooted at A (shortest path). Each node sends the packet to all its neighbors (besides the origin) All nodes should mark the transmitter of the first packet they receive as their parent on the tree. Nodes should relay the packet to their neighbors only once all subsequent receptions of the same packet should be ignored Now the node know where to send the packet addressed to the root. How should the root reach the node? Source routing...or tables... EECS SPRING of 5

31 Beyond trees Trees are not the best way to travel between any pair of nodes. Compare the way from E to A in the previous slide... We are looking for solutions assuring the shortest route between any to nodes There are numerous solutions... EECS SPRING 007 of 5

32 Classes of routing algorithms [steenkiste] Fully distributed porcessingseparately in each node but full graph structure needed in each node!. Fully distributed processing- no whole structure of The graph per node Bellmann-Ford EECS SPRING 007 of 5

33 Dijkstra Every node knows the graph All link weights are >= 0 Goal at node : Find the shortest paths from to all the other nodes. Each node computes the same shortest paths so they all agree on the routes 6 5 Strategy at node : Find the shortest paths in order of increasing path length EECS SPRING 007 of 5

34 Dijkstra intuition [steenkiste] EECS SPRING 007 of 5

35 Dijkstra Notation c(i,j) >=0 :cost of link from (i,j) D(,i): Shortest path from to i. D(,x,i): Shortest path from to i via x Let P(k) be the set of nodes k-closest to D(,5)= D(,6,5)=5 P()={,} 6 5 IDEA: Given P(k) we can find P(k+) efficiently: To get P(k+), observe that. This node cannot be in P(k). It must be one hop away from some node in P(k) Suppose were false. We picked i Node i has no edge into P(k) There must be a node x, not in P(k) such that x is one hop away from P(k) and D(,i)=D(,x)+D(x,i) But then, D(,x) < D(,i) and we would have picked x instead. Pick node(s) that is one hop away from P(k) that is closest to. Keep iterating until all nodes are in P EECS SPRING of 5

36 Dijkstra P()={,} D(,)= P()={,,5} D(,5)= P()={,,,5,6} D(,)= D(,6)= EECS SPRING of 5

37 Dijkstra - Forwarding Table At node 5 Outgoing Cost EECS SPRING of 5

38 Bellman-Ford Principle EECS SPRING of 5

39 Bellman-Ford Communicate current distance estimates of node to every other node This is called its distance vector: D i = (D(i,),D(i,),,D(i,n)) Initially, assume that all distance estimates are c(i,j) The nodes do not need to learn the entire topology Just the distance estimates (vectors) of their neighbors Periodically each node sends its distance vector to all of its neighbors At node : D : (0,,,,,) Neighbor estimates : (,0,,,, ) 6: (,,,,,0) 6 5 At node 6: D 6 : (,,,,,0) Neighbor estimates : (0,,,,,) 5: (,,,,0,) EECS SPRING of 5

40 Bellman-Ford Upon receiving a more recent distance vector from its neighbors, a node, i, stores it and revises D i : New D(i,d): The total cost to send it via neighbor j is the sum of The link cost c(i,j) The stored estimate to reach d from j Pick the lowest sum over all the neighbors D(i,d) = min jεn(i) {c(i,j) + D(j,d)} Old Info: DV: (,,,,,0) Neighbor estimates: : (0,,,,,) 5: (,,,,0,) 6 5 Node 6 receives (0,,,,,,) from (,,,,0,) from 5 Revised: DV: (,,,5,,0) Neighbor estimates: : (0,,,,,) 5: (,,,,0,) Send all packets to via. EECS SPRING of 5

41 Bellman-Ford Forwarding Table at 6 Estimates DV: (,,,,,0) Neighbor estimates: : (0,,,5,,) 5: (,,,,0,) 6 5 Node Cost EECS SPRING 007 of 5

42 Why does this compute shortest paths? Suppose in every tick each node sends its distance vector. Assume that initial distances are At time h, node i has as an estimate of the shortest path to node j that has <= h+ hops! D h+ (i,j) = min kεn(i) {D h (k) + c(i,k)} EECS SPRING 007 of 5

43 Asynchronous Bellman Ford In general, nodes are using different and possibly inconsistent estimates If no link changes after some time t, the algorithm will eventually converge to the shortest path No synchronization required at all TOC IP Routing Types DV EECS Asynchronous SPRING 007 of 5

44 Bellman-Ford Algorithm () EECS SPRING 007 of 5

45 Bad News Travels Slowly M D(,)=, D(,)=, D(,)= EECS SPRING of 5

46 Bad News Travels Slowly Node takes about M Iterations to figure out that D(,)=M M Fundamental Cause: After a network change, think of the network protocol running from time 0. The initial conditions are arbitrary Tricks exist to get around these problems but not fool proof EECS SPRING of 5

47 Oscillations Link costs must reflect link speed AND congestion Under both LSP and DV routing occurs over a tree The costs of the links of this tree will increase The other links will not be congested Their costs will drop Routing protocol will shift traffic and create a new tree This process of shifting and re-shifting can be severe Way out: Change congestion costs slowly (exponential averaging) Route dampening EECS SPRING of 5

48 Comparison: No clear winner Dijkstra: is robust since it each node computes its own routes independently Suffers from the weaknesses of the topology update protocol. Inconsistency etc. Excellent choice for a well engineered network within one administrative domain Bellman-Ford works well when the network is large since it requires no synchronization and has a trivial topology update algorithm Suffers from convergence delays Very simple to implement at each node Excellent choice for large networks EECS SPRING of 5