INTERNETWORKING - 1

INTERNETWORKING

Until now, we have implicitly assumed that there is a single homogeneous network, with each machine using the same protocol in each layer. Unfortunately, this assumption is wildly optimistic. Many different networks exist, including PANs, LANs, MANs, and WANs. We have described Ethernet, Internet over cable, the fixed and mobile telephone networks, 802.11, 802.16, and more. Numerous protocols are in widespread use across these networks in every layer. In the following sections, we will take a careful look at the issues that arise when two or more networks are connected to form an internetwork, or more simply an internet.

It would be much simpler to join networks together if everyone used a single networking technology, and it is often the case that there is a dominant kind of network, such as Ethernet. Some pundits speculate that the multiplicity of technologies will go away as soon as everyone realizes how wonderful [fill in your favorite network] is. Do not count on it. History shows this to be wishful thinking. Different kinds of networks grapple with different problems, so, for example, Ethernet and satellite networks are always likely to differ. Reusing existing systems, such as running data networks on top of cable, the telephone network, and power lines, adds constraints that cause the features of the networks to diverge. Heterogeneity is here to stay.

If there will always be different networks, it would be simpler if we did not need to interconnect them. This also is unlikely. Bob Metcalfe postulated that the value of a network with N nodes is the number of connections that may be made between the nodes, or N2 (Gilder, 1993). This means that large networks are much more valuable than small networks because they allow many more connections, so there always will be an incentive to combine smaller networks.

The Internet is the prime example of this interconnection. (We will write Internet with a capital ‘‘I’’ to distinguish it from other internets, or connected networks.) The purpose of joining all these networks is to allow users on any of them to communicate with users on all the other ones. When you pay an ISP for Internet service, you may be charged depending on the bandwidth of your line, but what you are really paying for is the ability to exchange packets with any other host that is also connected to the Internet. After all, the Internet would not be very popular if you could only send packets to other hosts in the same city.

Since networks often differ in important ways, getting packets from one network to another is not always so easy. We must address problems of heterogeneity, and also problems of scale as the resulting internet grows very large. We will begin by looking at how networks can differ to see what we are up against. Then we shall see the approach used so successfully by IP (Internet Protocol), the network layer protocol of the Internet, including techniques for tunneling through networks, routing in internetworks, and packet fragmentation.

How Networks Differ

Networks can differ in many ways. Some of the differences, such as different modulation techniques or frame formats, are internal to the physical and data link layers. These differences will not concern us here. Instead, in Fig. 5-38 we list some of the differences that can be exposed to the network layer. It is papering over these differences that makes internetworking more difficult than operating within a single network.

When packets sent by a source on one network must transit one or more foreign networks before reaching the destination network, many problems can occur at the interfaces between networks. To start with, the source needs to be able to address the destination. What do we do if the source is on an Ethernet network and the destination is on a WiMAX network? Assuming we can even specify a WiMAX destination from an Ethernet network, packets would cross from a connectionless network to a connection-oriented one. This may require that a new connection be set up on short notice, which injects a delay, and much overhead if the connection is not used for many more packets.

Many specific differences may have to be accommodated as well. How do we multicast a packet to a group with some members on a network that does not 

support multicast? The differing max packet sizes used by different networks can be a major nuisance, too. How do you pass an 8000-byte packet through a network whose maximum size is 1500 bytes? If packets on a connection-oriented network transit a connectionless network, they may arrive in a different order than they were sent. That is something the sender likely did not expect, and it might come as an (unpleasant) surprise to the receiver as well.

These kinds of differences can be papered over, with some effort. For example, a gateway joining two networks might generate separate packets for each destination in lieu of better network support for multicasting. A large packet might be broken up, sent in pieces, and then joined back together. Receivers might buffer packets and deliver them in order.

Networks also can differ in large respects that are more difficult to reconcile. The clearest example is quality of service. If one network has strong QoS and the other offers best effort service, it will be impossible to make bandwidth and delay guarantees for real-time traffic end to end. In fact, they can likely only be made while the best-effort network is operated at a low utilization, or hardly used, which is unlikely to be the goal of most ISPs. Security mechanisms are problematic, but at least encryption for confidentiality and data integrity can be layered on top of networks that do not already include it. Finally, differences in accounting can lead to unwelcome bills when normal usage suddenly becomes expensive, as roaming mobile phone users with data plans have discovered.

How Networks Can Be Connected

There are two basic choices for connecting different networks: we can build devices that translate or convert packets from each kind of network into packets for each other network, or, like good computer scientists, we can try to solve the problem by adding a layer of indirection and building a common layer on top of the different networks. In either case, the devices are placed at the boundaries between networks.

 Early on, Cerf and Kahn (1974) argued for a common layer to hide the differences of existing networks. This approach has been tremendously successful, and the layer they proposed was eventually separated into the TCP and IP protocols. Almost four decades later, IP is the foundation of the modern Internet. For this accomplishment, Cerf and Kahn were awarded the 2004 Turing Award, informally known as the Nobel Prize of computer science. IP provides a universal packet format that all routers recognize and that can be passed through almost every network. IP has extended its reach from computer networks to take over the telephone network. It also runs on sensor networks and other tiny devices that were once presumed too resource-constrained to support it.

We have discussed several different devices that connect networks, including repeaters, hubs, switches, bridges, routers, and gateways. Repeaters and hubs just move bits from one wire to another. They are mostly analog devices and do not understand anything about higher layer protocols. Bridges and switches operate at the link layer. They can be used to build networks, but only with minor protocol translation in the process, for example, between 10, 100 and 1000 Mbps Ethernet switches. Our focus in this section is interconnection devices that operate at the network layer, namely the routers. We will leave gateways, which are higherlayer interconnection devices, until later.

Let us first explore at a high level how interconnection with a common network layer can be used to interconnect dissimilar networks. An internet comprised of 802.11, MPLS, and Ethernet networks is shown in Fig. 5-39(a). Suppose that the source machine on the 802.11 network wants to send a packet to the destination machine on the Ethernet network. Since these technologies are different, and they are further separated by another kind of network (MPLS), some added processing is needed at the boundaries between the networks.

Because different networks may, in general, have different forms of addressing, the packet carries a network layer address that can identify any host across the three networks. The first boundary the packet reaches is when it transitions from an 802.11 network to an MPLS network. 802.11 provides a connectionless service, but MPLS provides a connection-oriented service. This means that a virtual circuit must be set up to cross that network. Once the packet has traveled along the virtual circuit, it will reach the Ethernet network. At this boundary, the packet may be too large to be carried, since 802.11 can work with larger frames than Ethernet. To handle this problem, the packet is divided into fragments, and each fragment is sent separately. When the fragments reach the destination, they are reassembled. Then the packet has completed its journey.

The protocol processing for this journey is shown in Fig. 5-39(b). The source accepts data from the transport layer and generates a packet with the common network layer header, which is IP in this example. The network header contains the

ultimate destination address, which is used to determine that the packet should be sent via the first router. So the packet is encapsulated in an 802.11 frame whose destination is the first router and transmitted. At the router, the packet is removed from the frame’s data field and the 802.11 frame header is discarded. The router now examines the IP address in the packet and looks up this address in its routing table. Based on this address, it decides to send the packet to the second router next. For this part of the path, an MPLS virtual circuit must be established to the second router and the packet must be encapsulated with MPLS headers that travel this circuit. At the far end, the MPLS header is discarded and the network address is again consulted to find the next network layer hop. It is the destination itself. Since the packet is too long to be sent over Ethernet, it is split into two portions. Each of these portions is put into the data field of an Ethernet frame and sent to the Ethernet address of the destination. At the destination, the Ethernet header is stripped from each of the frames, and the contents are reassembled. The packet has finally reached its destination.

Observe that there is an essential difference between the routed case and the switched (or bridged) case. With a router, the packet is extracted from the frame and the network address in the packet is used for deciding where to send it. With a switch (or bridge), the entire frame is transported on the basis of its MAC address. Switches do not have to understand the network layer protocol being used to switch packets. Routers do.

Unfortunately, internetworking is not as easy as we have made it sound. In fact, when bridges were introduced, it was intended that they would join different types of networks, or at least different types of LANs. They were to do this by translating frames from one LAN into frames from another LAN. However, this did not work well, for the same reason that internetworking is difficult: the differences in the features of LANs, such as different maximum packet sizes and LANs with and without priority classes, are hard to mask. Today, bridges are predominantly used to connect the same kind of network at the link layer, and routers connect different networks at the network layer.

Internetworking has been very successful at building large networks, but it only works when there is a common network layer. There have, in fact, been many network protocols over time. Getting everybody to agree on a single format is difficult when companies perceive it to their commercial advantage to have a proprietary format that they control. Examples besides IP, which is now the near-universal network protocol, were IPX, SNA, and AppleTalk. None of these protocols are still in widespread use, but there will always be other protocols. The most relevant example now is probably IPv4 and IPv6. While these are both versions of IP, they are not compatible (or it would not have been necessary to create IPv6).

A router that can handle multiple network protocols is called a multiprotocol router. It must either translate the protocols, or leave connection for a higher protocol layer. Neither approach is entirely satisfactory. Connection at a higher layer, say, by using TCP, requires that all the networks implement TCP (which may not be the case). Then, it limits usage across the networks to applications that use TCP (which does not include many real-time applications).

The alternative is to translate packets between the networks. However, unless the packet formats are close relatives with the same information fields, such conversions will always be incomplete and often doomed to failure. For example, IPv6 addresses are 128 bits long. They will not fit in a 32-bit IPv4 address field, no matter how hard the router tries. Getting IPv4 and IPv6 to run in the same network has proven to be a major obstacle to the deployment of IPv6. (To be fair, so has getting customers to understand why they should want IPv6 in the first place.) Greater problems can be expected when translating between fundamentally different protocols, such as connectionless and connection-oriented network protocols. Given these difficulties, conversion is only rarely attempted. Arguably, even IP has only worked so well by serving as a kind of lowest common denominator. It requires little of the networks on which it runs, but offers only best-effort service as a result.

Frequently Asked Questions