The Internet

The hype surrounding mobile data is ultimately founded on one thing: the Internet. Vendors and operators alike use slogans such as "Internet anywhere" and "Internet in your pocket," promising to cut the Internet free from its PC-based roots. Trouble is, no matter how advanced the systems sending data over the air, the Internet itself may not be up to the job.

The Internet was founded as an experiment, linking together a few computers involved in defense research. It was never intended to become a communications system for the masses, much less the basis of the world economy. Exponential growth in fixed access is already pushing the network to its limits, and connecting every mobile phone and PDA might be just enough to knock it over the edge.

The problem lies in the addressing scheme that the Internet uses to identify computers and ensure that each one gets the correct data. While fiber-optic cables and more powerful computer chips mean that the amount of data passing through the Internet can grow without limit, the number of computers linked to it and the complexity of the network cannot.

Internet Protocol Version 4

Each computer, or node, connected to the Internet is identified by a unique number, known as the IP address. This must be encoded into every packet of data that crosses the network, where it is read by routers, machines that forward data packets via the most efficient path. The present version of IP represents this address with a 32-bit number, usually written as four eight-bit bytes. For this reason, it is known as IP v4 (there were no versions 1 to 3).

In theory, there are 2³² or 4,294,967,296 possible addresses, but no router can learn the path to every one of them. Instead, routers at the Internet core only forward packets based on the initial bits of the address, known as the network prefix. As all addresses have 32 bits, a shorter prefix means a larger network, and vice versa. The number of machines on each network needs to be a power of two, which on average would waste a quarter of all addresses if every network took the smallest allocation mathematically possible. For example, a network with 100 nodes would need 128 (or 2⁷) addresses, while one with 300 would need 512 (or 2⁹).

Unfortunately, there are further practical limitations. First, most routers at the Internet core do not recognize networks smaller than 4,096 (or 2¹²) machines, because these require long prefixes and each extra bit in the prefix doubles a router's memory requirement. Second, the original IP v4 specification allowed only three different sizes of networks. They were known as classes A, B, or C, and used 8-bit, 16-bit, or 32-bit network prefixes respectively. Routers were programmed to determine the class by looking at the first four bits of the address, as shown in Table 8.1 for example, the largest networks were class A, whose addresses always began with a 0. There were also two extra classes, D and E, used for broadcast traffic and for testing.

The class scheme meant that users requiring only a few hundred addresses had to request thousands or even millions. Early adopters were duly handed out class A and B networks, which most still hold on to. In 1996, IANA (Internet Assigned Numbers Authority) issued an appeal to return unused address space, which has largely been ignored. Only one complete class A has actually been returned: network 36, which until May 2000 belonged to Stanford University, one of the Internet's founders. The cybersquatters still holding onto other class A's include some of the world's largest corporations and most renowned academic institutions.

Table 8.1. IP v4 Address Classes

Class	Address Range	Initial Bits	Prefix size	Nodes Per Network	Number of Networks	Proportion of Internet
A	0.0.0.0 - 127.255.255.255	0xxx	/8	1,677,7216	128	50
B	128.0.0.0 - 191.255.255.255	10xx	/16	65536	16,384	25%
C	192.0.0.0 - 223.255.255.255	110x	/24	256	2,097,152	12.5%
D	224.0.0.0 - 239.255.255.255	1110	/4	268,435,456	1	6.25%
E	240.0.0.0 - 255.255.255.255	1111	/4	268,435,456	1	6.25%

The IP Address Shortage

In the twentieth century, many businesses and almost all home users only connected to the Internet for short periods of time. This meant that they could share IP addresses with many others, requesting a temporary address when connecting and then returning it when they disconnected. An ISP needed only as many addresses as it had modems, not as many as it had users.

This is quickly changing. Many U.S. consumers are already seeing the benefits of fast, always-on connections via DSL (Digital Subscriber Line), the fixed access technology that achieves high data speeds over ordinary telephone lines. But what really threatens the IP address pool is GRPS, the always-on technology added to mobile phones. Though its capacity is far lower than that of DSL, a GPRS phone still requires a permanent IP address.

In March 2000, the GSM Association requested two class A's, nearly one percent of the entire IP v4 address space. This may not sound like much, but it's just the beginning. The problem is particularly severe in Europe, where mobile phones are very popular but IP address allocations have traditionally been low, thanks to the Internet's American origins. RIPE (Réseaux IP Européens), the regional registry that handles Europe, has control over a space covering only ten million addresses, while U.S. users have been allocated around two billion. RIPE simply can't afford to give away its address space to European mobile operators.

Worldwide, wireless vendors boast that mobile phones are giving many people the opportunity to make their first ever phone call. Mobile networks are easier than fixed to set up in remote areas, and falling costs mean that telecom will soon be within reach of some of the world's poorest people. But with a global population of more than six billion and all new phones Internet-equipped, there won't be enough addresses to go round.

Mobile IP

Because of the hierarchical nature of the Internet, a user can't just take her computer and its IP address wherever she goes. IP addresses are specific to whatever network the prefix defines, and won't work outside of it. Otherwise, core routers would need to remember all four billion addresses and their associated routes.

To get around this, the Mobile IP standard was published in 1996. It was designed for world travelers using laptops over fixed networks, but has been adopted by the wireless world. It uses a system called tunneling, which requires users to adopt a second, temporary IP address whenever they connect to networks other than the one to which their own IP address belongs. The home network then routes packets intended for the user to this temporary address, rather like having mail forwarded from your home while on a long vacation.

The problem with Mobile IP is that it routes all incoming packets via the user's home network, causing the same inefficiencies as roaming between cellphone networks. It isn't as noticeable to the user because Internet usage isn't billed by distance, but it does waste capacity and add latency. It also wastes IP addresses, as the user is temporarily given two.

Outgoing data packets should not have to be routed via the user's home network, just as most vacationers would not send a postcard via their home address. But security considerations mean that they often are sent this way, further reducing performance. Every packet has to contain the sender's IP address, which for a mobile user will not match the network they are actually in. Many firewalls reject such traffic because it is often a sign that the address has been forged, a common trick among hackers.

Internet Protocol Version 6

As a permanent solution to the address shortage, network visionaries since the mid-1990s have been pushing a new IP standard, which uses 128-bit addresses. These split into 16 bytes, so the new standard is known as IP v6. Computer engineers often say 6 when they mean 16; Intel did the same with its x86 processors.

IP v6 allows a total of more than 342,000 decillion (or 2¹²⁸) addresses, which is enough to give every atom in the Universe its own Internet connection. The new standard also includes advanced quality of service features and new ways of dealing with mobility. Unfortunately, it is still in the future. Every network architect agrees it is great in theory, but doesn't want to use it in practice. According to ARIN (American Registry for Internet Numbers), which issues IP addresses in America, more than 800 new networks requested blocks of IP v4 space in 1999, compared with only two requests for IP v6.

normal: Web Resources http://www.wapgateway.org This is the site of a Finnish project building an open-source (free) WAP Gateway. The code can be downloaded from the site and will run on a standard PC under the open-source operating system Linux. http://www.iana.org The Internet Assigned Numbers Authority is in overall charge of the IP address space. Users requesting addresses for a network should go straight to the regional registries linked to from its site. http://www.cis.ohio-state.edu/Services/rfc/ RFCs (Requests For Comment) are the Internet's official standards and contain guidelines on every aspect of Internet engineering. They can all be downloaded free of charge from many Websites and copied without permission or royalty. The most relevant to this section are RFC 791 (Internet Protocol v4), RFC 1366 (Address Space Management), RFC 1518 (Classless Routing), RFC 1917 (Appeal to Return Unused Networks), RFC 2002 (Mobile IP), and RFC 2460 (Internet Protocol v6). http://www.protocols.com This site, run by Israeli equipment manufacturer RAD, details all common protocols, from IP down to the bit level. http://www.base-earth.com The magazine Base Station Earth Station covers the equipment needed to build a mobile network and the issues faced by cellular engineers.