Telecommunication Protocols Laboratory Course. Lecture 3

Transcription

1 Telecommunication Protocols Laboratory Course Lecture 3

2 Course map Last time: we discussed protocols of the Medium Access Control (MAC) sub-layer Deal with broadcast channels and their (multi-party) protocols Key issue: how to determine who gets to use the channel when there is competition for it MAC is a sub-layer of DLL Today: we go one layer up, to NETWORK LAYER (NL) We study routing algorithms and NL protocols Focus on NL in Internet April 5,

3 Network layer Role: to get packets from source all the way to the destination This may require many hops at intermediate routers on the way Lowest layer dealing with end-to-end transmission DLL was moving frames from one end of the wire to the other To implement its role, NL has to Know the topology of the communication subnet (all routers) Choose appropriate paths through it, perhaps to avoid congestion Dealing with possibly different networks for source and destination April 5,

4 Network Layer Design Issues Services provided to the transport layer Internal design of the subnet April 5,

5 Context in which NL protocols operate Carrier s equipment routers connected by transmission lines Customer s equipment hosts and possibly also routers Leased lines from host/customer router to carrier s equipment

6 Store-and-Forward Packet Switching Mechanism Equipment used as follows A host with a packet sends it to its nearest router Packet stored until it has fully arrived, so checksum can be verified Then forwarded to the next router along the path until destination host is reached There it is delivered Forwarding done based on some local information kept in routing tables Routing algorithms: manage the tables and make routing decisions April 5,

7 Services provided to TL NL services designed with following goals Services should be independent of the router technology TL should be shielded from the number, type, topology of present routers The network address made available to TL should use a uniform numbering plan Lots of freedom in writing detailed specifications of the services to be offered to TL battle between connectionless and connection-oriented services April 5,

8 Connectionless services Advocated by Internet community Experience: 30 years with a real, working computer network Routers job is only to move packets around Subnet inherently unreliable Hosts should accept this and do error control and flow control Network service should be connectionless, having primitives SEND PACKET and RECEIVE PACKET No packet ordering or flow control since hosts do that Each packet must carry the full destination address April 5,

9 Connectionless service implementation Packets injected into subnet individually and routed independently of the others No advance setup needed Packets datagrams (analogy with telegrams) Subnet datagram subnet April 5,

10 Implementation of Connectionless Service Routing within a diagram subnet. April 5,

11 Connection-oriented services Advocated by telephone companies Experience: 100 years success with world-wide telephone systems Subnet should provide a reliable, connection-oriented service Quality of service is the dominant factor Hard to achieve w/o connections in subnet, especially for real-time traffic such as voice and video April 5,

12 Connection-oriented service implementation Path from source S to destination D has to be established before data packets are sent Path virtual circuit (VC) (analogy with physical circuits set up in telephone systems) Subnet virtual circuit subnet Idea: avoid choosing a new route for each packet When connection established Route from S to D chosen as part of the communication setup Route stored in router tables and used for all traffic flowing in connection Each packet carries an identifier telling which VC it belongs to When connection released, VC terminated

13 Implementation of Connection-Oriented Service Routing within a virtual-circuit subnet.

14 Virtual-Circuit and Datagram Subnets 5-4

15 Routing algorithm Chooses the routes and data structures to use for routing packets from S to D Differently expressed: that part of NL software responsible for deciding which output line an incoming packet should be transmitted on Routing and forwarding processes inside router FWD handles each packet as it arrives, looking up the outgoing line to use for it in routing tables ROUT responsible for filling in and updating the routing tables April 5,

16 Desired properties of the routing algorithms Correctness Simplicity Stability Robustness Fairness Optimality April 5,

17 Conflict between fairness and optimality April 5,

18 Types of algorithms Non-adaptive (static) Do not consider measurements/estimates of current traffic or topology in their routing decisions The route from I to J ( I, J) computed in advance, off-line and downloaded when network booted Adaptive (dynamic) Routing decisions change to reflect changes in topology and traffic These algorithms differ in Where they get their information (locally, from adjacent/all routers) When they change routes (every t sec, when load/topology changes) What metric is used for optimization (distance, hops nr, transit time) April 5,

19 Optimality principle General statement about optimal routes without regard to network topology or traffic If router J is on the optimal path from I to K (routers both), then the optimal path from J to K falls along same route The set of optimal routes from all sources to a given destination form a tree routed at destination: SINK Tree Goal of all routing algorithms: find and use sink tree The principle and the sink tree benchmark against which other routing algorithms can be measured April 5,

20 The Optimality Principle (a) A subnet. (b) A sink tree for router B. April 5,

21 Routing Algorithms Shortest Path Routing Flooding Distance Vector Routing Link State Routing Hierarchical Routing Broadcast Routing April 5,

22 Shortest Path Routing Non-adaptive algorithm Idea: build a graph of the subnet, so that: router node, communication line arc Algorithm finds shortest path between S and D in the graph, with respect to Nr of hops Geographic distance Mean queuing and transmission delay Labels could also be computed as function of the above April 5,

23 Shortest Path Routing example The first 5 steps used in computing the shortest path from A to D. The arrows indicate the working node.

24 Flooding algorithm Non-adaptive algorithm Every incoming packet is sent out on every outgoing line except the one it arrived on Vast nr of duplicate packets Some damping measures needed April 5,

25 Damping measures in flooding Hop counter in packet header, decremented at every hop, packet discarded when counter=0 Counter initially has the length of path from S to D, or diameter of subnet Sequence number in packet header given by router S Each router needs a list per router to know which numbers from that source already passed If packet has the number on the list, then it is not flooded further Simplification: only a counter k kept and all numbers < k are not flooded Selective flooding: only packets going right are flooded April 5,

26 Flooding usage Military applications Large nr of routers may be blown to bits at any moment Distributed database applications All tables need to get updates concurrently Wireless networks Broadcast properties of radio range is flooding As a metric against which other routing algorithms can be measured Shortest path as every possible path is chosen in parallel => no shorter delay possible April 5,

27 Distance Vector Routing Adaptive algorithm Original ARPANET routing algorithm, used also in Internet as RIP Each router maintains a table (vector) giving The best known distance to each destination Which line to use to get there Table updated by exchanging information with neighbours April 5,

28 Example for calculating the table Assume the metric is delay Assume delay to neighbors known Can be found out with special ECHO packets Once every T msec each router Sends to all neighbors a list of its estimated delays to each destination Receives similar list from neighbors Assume neighbor X has x i delay to get to router i If our router has m delay to get to X => m+x i to get to i It compares all calculations and registers the best one in its new routing table Note: new table is not based on old table April 5,

29 Distance Vector Routing example (a) A subnet. (b) Input from A, I, H, K, and the new routing table for J.

30 Efficiency of distance vector routing It converges to the correct answer but slowly, especially for bad news Known as count-to-infinity problem Attempts to solve it do not work well in general Core of the problem When X tells Y it has a path somewhere, Y has no idea if it itself (Y) is on that path April 5,

31 The count-to-infinity problem April 5,

32 Link State Routing In 79 it replaced distance vector routing in ARPANET Variants used widely today Each router must do the following Discover its neighbors, learn their network address. Measure the delay or cost to each of its neighbors. Construct a packet telling all it has just learned. Send this packet to all other routers. Compute the shortest path to every other router. All delays and the complete topology are experimentally measured and distributed to every router April 5,

33 Learning about the Neighbors When a router is booted It sends a special HELLO packet on each p2p line Router on the other end replies with its address Addresses must be unique April 5,

34 Measuring Line Cost Special ECHO packet sent; the other side requested to reply immediately The round trip is measured and split to 2 => reliable estimate If test conducted several times, average is used This method implies delays are symmetric April 5,

35 Building Link State Packets Each packet has Identity of sender Sequential nr Age List of neighbors, with address and delay Building packets is easy When to build them? Periodically When some significant event occurs (line/neighbor goes down/ comes back or changes properties significantly) April 5,

36 Example of Link State Packets (a) A subnet. (b) The link state packets for this subnet. April 5,

37 Distributing the Link State Packets Trickiest part of algorithm distributing packets reliably Flooding used to distribute each link state packet Sequence numbers used, increased for each new packet sent Routers keep lists of all (source-router, seq-nr) seen: when a packet comes in, it is checked against the lists If new: flooded If old: discarded (already seen or not seen but obsolete) Some problems: when seq. nr. wrap around, when router crashes, when seq. nr. corrupted April 5,

38 Distributing the Link State Packets Age number used (more) Decrement it once per sec; when age=0, packet discarded Decremented also by routers during initial flooding process Refinements: holding area (h-area) When packet P arrives at router to be flooded it is stored shortly in h-area If another packet P comes from same source before P is transmitted, compare seq-nr: if equal then duplicate discarded, if distinct then older one thrown out All packets acknowledged When line is idle, h-area scanned to select packet to send April 5,

39 Computing the new routes Router has now a full set of link state packets so it can construct the subnet graph Each link represented twice, once for each direction These two values can be averaged or used separately Dijkstra s algorithm used (run locally) to construct shortest path to all possible destinations Results of algorithm installed in routing tables and normal operation resumed April 5,

40 Hierarchical Routing Networks grow in size Router routing tables grow proportionally Memory consumed by increasing tables More CPU time needed to scan them More bandwidth needed for sending status reports about them It may not be feasible for every router to have an entry for each other router in large networks Routing should be done hierarchically, like in telephone networks April 5,

41 Hierarchical Routing (more) Routers divided into regions Each router knows all details about routing packets to destinations within its own region Many levels of hierarchy might be needed Saving in table space might mean some longer paths How many levels? N routers, lnn levels, with elnn entries/router ( 79) This result also means increase in path length are ok April 5,

42 Hierarchical Routing example April 5,

43 Broadcast Routing Some applications have hosts needing to send to many/all other hosts Service distributing weather reports/stock market updates Live radio programs Broadcasting to all machines is used Those interested read the data Many methods April 5,

44 Methods of broadcast routing Point-to-point (p2p): source sends a distinct packet to all destinations No special features required from subnet Bandwidth wasteful Source has to keep complete list of all destinations Least desirable Flooding Ill-suited for p2p communications Might be considered if other methods do not work Too many packets, too much bandwidth April 5,

45 Methods of broadcast routing (more) Multi-destination routing Each packet has list of destinations/bitmap indicating desired destinations At a router, destinations checked to see which out-lines needed Router generates new copy of packet for each out-line, including in packet only those destinations that use that out-line => destinations partitioned among out-lines After a sufficient nr of hops, each packet contains only one destination, so can be treated as a normal packet Spanning tree (ST) subset of subnet with all routers and no loop Ifeach router knows which of its lines belong to ST, then it can copy an incoming packet to all ST lines except the one it arrived on Excellent use of bandwidth, optimal approach Each router has to know of some spanning tree April 5,

46 Methods of broadcast routing 3 Reverse path forwarding Attempt to approximate behaviour of previous one, with routers knowing nothing of ST When packet arrives at router, router checks to see if it s on the line normally used to send packets to that source If so, excellent chance that packet followed that route and hence is the first copy to arrive => router forwards copies to all out-lines except the one it arrived on If not, packet discarded as a likely duplicate Reasonably efficient and easy to implement April 5,

47 The Network Layer in the Internet The IP Protocol IP Addresses Internet Control Protocols April 5,

48 Internet: Collection of Subnetworks

49 Communication in Internet TL takes data streams and breaks them up into datagrams Datagrams transmitted through Internet, possibly fragmented When all pieces get to destination, they are reassembled by NL into original datagram Datagram handled to TL which inserts it into the input stream of receiving process April 5,

50 IP (Internet Protocol) Glue that holds Internet together Designed from the beginning with internetworking in mind Provides a best-efforts way to transport datagrams from S to D, without regard whether these machines are on the same network or there are other networks in between April 5,

51 The IP Protocol IP datagram: text part and header part. Above, the IPv4 header. April 5,

52 IPv4 header options Intended as escape for subsequent versions of protocol Have variable length, each begins with a 1-byte code identifying the option Some options are followed by another byte telling the length April 5,

53 IP Addresses 32-bit addresses, each host/router has one Encode network nr and host nr Used in source and destination fields of IP packets IP address does not refer to a host but to a network interface If a host is on 2 networks, should have 2 IPs April 5,

54 Classes for IP Addresses April 5,

55 Special IP Addresses Special IP addresses. April 5,

56 Subnets Problem A single class A/B/C address refers to one network, not a collection of LANs; LANs cannot physically extend too much Solution Network split into several parts (subnets) for internal use so that they still act as one happy network to the outside world How to know where (which subnet) to send an incoming packet? (Table with entries in main router mapping router to host) Addressing convention April 5,

57 Subnets illustration A campus network consisting of LANs for various departments. April 5,

58 Example subnet Class B addresses: 14 bits network nr, 16 bits host nr For a subnet: 14 bits network nr 6 bits subnet nr 10 bits host nr 64 subnets, each with max 1022 hosts This split can be later changed if not ok Main router needs subnet mask to indicate the split: o /22 (22 nr of bits of network + subnet part) April 5,

59 The subnet mask example A class B network sub-netted into 64 subnets. April 5,

60 Internet Control Protocols IP used for data transfer Several control protocols exist ICMP, ARP, RARP, BOOTP, DHCP ICMP (Internet Control Message Protocol) Operation of Internet monitored closely by routers When something unexpected occurs, event is reported by ICMP Also used to test the Internet Types of ICMP messages are defined April 5,

61 The principal ICMP message types 5-61 April 5,

62 ARP Address resolution protocol Problem: IP addresses cannot be used to send packets as DLL hardware does not recognize them Solution: some kind of mapping is needed In ARP Broadcast packet asking who owns IP x? All machines get the packet, check their IP and one host replies => Sender builds Ethernet frame addressed to that host and puts IP packet in payload; receiver does the opposite April 5,

63 Optimization of ARP Cache Machines that had run ARP, cache the result in case it s needed again Sender can avoid another broadcast from receiver if it sends IP-to-Ethernet mapping in ARP packet All machines receiving the broadcast can cache it Boot Every machine broadcasts mapping when booting and everyone caches it Time out Entries in ARP cache should time out in few minutes, to allow for changes Proxy ARP One router can be configured to respond to ARP requests for local networks OR: sending hosts recognize they are sending to a remote host and send all such traffic to default Ethernet address handling all remote traffic April 5,

64 RARP Reverse ARP ARP: IP Ethernet; RARP: Ethernet IP Problem occurs when a diskless workstation WS is booted WS gets the binary image of its OS from remote file server Host asks what s my IP? (with a broadcast message: 11 1) RARP replies upon looking up the Ethernet address in its configuration files Better than embedding IP in binary image, as one image is thus used for all stations Disadvantage host broadcasts locally, so its message cannot be forwarded by routers => RARP server needed on each network April 5,

65 BOOTP Bootstrap protocol Gets around RARP disadvantage by using UDP messages, forwarded to routers Provides WS with additional information IP address of file server holding the memory image IP address of default router Subnet mask Problem Requires manual configuration of tables mapping IP Ethernet April 5,

66 DHCP Dynamic host configuration protocol Eliminates BOOTP s problem by allowing both manual and automatic IP addresses assignment Extension of BOOTP; largely replaces BOOTP and RARP DHCP server assigns IP addresses to hosts asking for one DHCP server may be on remote network and is not reachable by broadcasting DHCP relay agent needed on each LAN Newly booted machines broadcast DHCP DISCOVER packet DHCP relay agent sends packet as unicast to DHCP server Leasing: IP addresses allocated for a finite time April 5,

67 Operation of DHCP April 5,

68 NL simulation You may choose either to simulate a routing algorithm or a NL protocol. Choose from: Routing algorithms Shortest Path Routing Flooding Distance Vector Routing Link State Routing Hierarchical Routing Broadcast Routing NL protocols IP ICMP ARP RARP BOOTP DHCP April 5,