4.3 IP over ATM
You have seen that the IP protocol supports a connectionless service, and the ATM and TCP protocols support a connection-oriented service.
SAQ 8 (Revision)
What are the relative advantages and disadvantages of connectionless and connection-oriented services for transporting data through a communication network?
My list of the advantages and disadvantages of connectionless and connection-oriented services is given in Table 11. The list is not exhaustive and you may have included some other factors. Also you may wish to qualify some of those I have listed.
Table 11: Answer to SAQ 8 (Revision)
|Type of service||Advantages||Disadvantages|
|Connectionless||Data can be sent immediately without the overhead of setting up a connection||No intrinsic flow control No intrinsic error correction|
|It is robust against link failures||Forwarding is based on full destination address|
|The network switches do not need to maintain connections||Order of packets is not guaranteed|
|There can be wide variation in the delay|
|Connection- oriented||Flow control can be implemented within the service||Overhead of setting up a connection before data is sent|
|Error control can be implemented within the service||Switches have to maintain connections|
|Order of packets is maintained|
|Forwarding is done according to simple connection identifiers|
The Internet and internetworks in general consist of many physical networks, which may be of different types. After all, one of the main advantages of the TCP/IP protocol suite is its compatibility with many different network technologies. The host-to-network layer in the TCP/IP protocol model may transfer data over a range of network links – relatively slow modem links to homes, fast LANs interconnecting many computers within a building, fast private point-to-point links rented from telephone network operators. The example I wish to explore in this section is sending IP traffic over an ATM network.
Telephone network operators have installed a large amount of ATM equipment and it is likely that in the future a significant amount of traffic will be transported over ATM networks for part of its path between users. The ATM portions of the paths are of no interest to users because all they are concerned about is the quality of service they receive, not how it is achieved.
Several techniques exist for sending IP traffic over ATM networks. The technique I shall describe is label switching. However, you should bear in mind that there are other approaches and that label switching has wider application than supporting IP traffic over ATM networks. I shall ignore many of the intricate problems that arise when two significantly different technologies work together.
Before discussing label switching, however, I need to review the two basic mechanisms for transporting packets in packet-switched networks, which I shall call forwarding and routing. Unfortunately, these terms tend to be used interchangeably in the literature and their exact meaning has to be derived from the context. In this section I shall try to be more precise and define forwarding as the process of receiving a packet and passing it on to the next stage in the path to its destination, that is, packet switching. This generally involves a packet switch referring to a table that identifies the next path for each destination. This table is commonly called a forwarding table (see examples discussed earlier for ATM and IP), but it is also frequently called a routing table.
The entries in a forwarding table are determined by the routing process. The best path through a network is chosen from information about which links are available between switches. Selection of the best path can be based on the current level of traffic, which means that the contents of a forwarding table will change. Changes are made in the background to the forwarding process. Protocols that are specific to a network transport routing information between switches.
Generally in a large IP internetwork, it is impracticable for each router to contain an entry for each network address, so several network addresses are amalgamated into a single entry called & forwarding equivalence class. For example, imagine that a router knows that all packets destined for network addresses 188.8.131.52/16 (10000100 10001101), 184.108.40.206/16 (10000100 10011000) and 220.127.116.11/16 (10000100 10101100) should be forwarded over the same path. The 10 most significant bits of the 16-bit network addresses have the same value for all three addresses, so the forwarding table could have a single entry for the value 18.104.22.168/10 (10000100 10000000).
What is the longest forwarding equivalence class address if datagrams for the network addresses 22.214.171.124/8, 126.96.36.199/8 and 188.8.131.52/8 are forwarded over the same interface?
Converting the network addresses to binary values:
The longest match is on the first seven bits of the network addresses, which gives 200/7 or 201/7. The network addresses 200/7 and 201/7 are indistinguishable, so either value could be chosen.
With forwarding equivalence classes it is possible for a network address to match more than one entry in a forwarding table and the path chosen will be the one with the longest matching address.
The following network addresses appear in a forwarding table:
Which of the entries match the IPv4 address 184.108.40.206?
The IPv4 address as a binary number is 10110110 10010001 00111001 00000011 and comparing this address with the table entries gives:
|Table entry||Destination network address||Result|
|(b)||10110110 01||10110110 10||No|
|(c)||10110110 10||10110110 10||Yes|
|(d)||10110110 1010||10110110 1001||No|
|(e)||10110110 1001||10110110 1001||Yes|
The searching for the longest match in a forwarding table takes place in each router in a connectionless network. The forwarding decision in a connection-oriented network is based on virtual path identifiers, which tend to be of fixed lengths. This simplifies the searching process. Label switching in essence brings this advantage of connection-oriented networks to connectionless protocols. Within a label-switched network, packets are forwarded according to labels added to them rather than according to the destination address stored in the packet header.
A router that is located around the periphery, or boundary, of a network is called an edge router. At an edge router of a label-switched network the destination addresses have to be converted to labels, but once within a label-switched network forwarding is achieved by inspection of labels alone. In ATM networks the labels can be inserted in the existing fields for virtual path and/or virtual channel identifiers. In other networks it may be necessary to encapsulate the IP datagram in another packet which contains a label in its header.
Label switching seems to add extra overheads to forwarding packets. Suggest what advantages it may have.
Possible advantages include the following:
if the label-switched network extends for most of the path between end users, then most of the packet forwarding would be done more efficiently than with conventional routers;
label switching allows IP traffic to be carried by connection-oriented networks such as ATM;
it may be possible to incorporate greater control over the quality of service by assigning packets associated with particular applications to higher-quality virtual circuits.
Figure 30 shows the operation of IP over ATM using label switching. Although the figure shows only the entries for the paths from IP network 131.30.xx, it is a fairly complicated diagram, so you will need to spend some time working through the forwarding tables. The association of labels with IP forwarding equivalence class addresses is part of the routing process and is accomplished by a label distribution protocol in conjunction with the ATM signalling protocol for establishing virtual circuits.
Compare the functions of forwarding IP datagrams and ATM data cells.
IP datagrams are forwarded by routers (packet switches) according to network addresses and as a connectionless service. The path taken depends on the content of forwarding tables, which reflect the outcome of routing operations. The choice of path may vary during a flow of datagrams between users if the forwarding tables are updated.
ATM data cells are forwarded (switched) according to the values of their virtual path and virtual channel identifiers. These identifiers refer to either a permanent virtual path or a switched virtual path, but in both cases a connection-oriented service is provided to users.
The values of virtual path and virtual channel identifiers may vary as cells are forwarded (i.e. they have local significance), whereas network destination addresses in IP datagrams do not vary (i.e. they have global significance).