There were more than 93 million hosts connected to the Internet in July 2000, according to Mark Lotter's Internet Domain Survey. That number is growing annually at a rate of 63 percent, which means 69 new hosts and 23 new domains are created every minute.
Did somebody say 800-pound gorilla? And today's IP addressing standards just can't keep up with its appetite.
The current Internet protocol version four (IPv4) implements a 32-bit addressing scheme whose capacity caps at 4.3 billion addresses (232 bits = 4,294,967,296 available IPv4 addresses). With a world population of 6 billion, and only 59 million of those active Internet users, that may seem like plenty of addresses to go around.
But the required number of IP addresses does not map to the number of Internet users. Many users have more than one host, and a host is not necessarily associated with a human user. Every data link within the Internet (for example, each network interface on every router or IAD port) requires a unique address identifier to make the transmission and delivery of data possible.
To make matters worse, hungry addressing schemes and address hoarding have left as much as 40 percent of all allocated address space unused.
Obviously, something must be done. We must wean the network off its appetite for IP addresses.
Any method of improving the efficiency of IP addressing, and of reducing the address space required, is an advantage to a service provider's business.
Reduced management and administrative requirements translate into lower recurring and operational costs. Addressing innovations can enable more flexible, higher-value services such as dynamic address provisioning.
Most important, in the face of such problems as address hoarding, a service provider's ability to demonstrate responsible conservative addressing practices can assist in obtaining large address blocks--a clear benefit to ISP.
Unfortunately, IP addressing today--and its more recent enhancements--just won't fit the bill.
IPv4's addressing scheme is known as "classful" addressing because it constricts the distribution of addresses by grouping them into classes. IPv4 is divided into five address classes--A to E--each with a finite number of addresses designed for networks of particular sizes. Fixed bit length in the network identifier portion of the address determines the class.
For example, 50 percent (or 2,147,483,648) of IPv4 address space is dedicated to Class A addresses (called "eights", because they reserve eight bits for the network identifier). A maximum of 126 networks can be defined as eights. Each of those networks can accommodate 16.7 million hosts.
Likewise, there is space in IPv4 for more than 16,000 Class B networks, or 16s, of 65,534 hosts each; and 2.1 million Class C networks, or 24s, with up to 254 hosts each. Classes D and E are reserved address spaces that together consume 1/8 of the total space. Class D is reserved for IP multicasting, and Class E for experimenting and testing.
The major difficulties with classful addressing are related to Class B, the space that serves medium-sized organizations. In the past, Class Bs have been given to organizations that require more than a couple of Class Cs, thus rapidly depleting Class B space.
More problematic is what an organization is to do once its Class C network grows beyond the number of hosts in its initial allocation. The leap from a network size of 254 hosts to a Class B address of more than 16,000 hosts does not generally reflect the growth pattern of most organizations.
The good news is, the industry is on the case.
The bad news is, it will be on the case for some time to come.
IPv6 to the Rescue?
IP version six (IPv6 or IPng, for "next-generation") proposes an addressing scheme of 128 bits. Douglas Comer, author of Computer Networks and Internets, estimates that this address space is so large that "every person on the planet can have sufficient addresses to have their own internet as large as the current Internet."
IPv6 also includes impressive perks, such as automatic address configuration, mobile IP support for wireless routing, and "anycast" addresses, which reduce routing overhead currently caused by the forwarding of multicast messages to multiple addresses.
But IPv6 still poses daunting problems regarding transition and backward compatibility. Just think about it: The entire Internet has to be renumbered.
Simply put, the Internet cannot wait for IPv6 to be agreed upon and implemented in time.
Not surprisingly, stop-gap solutions have been developed to extend the address space offered by IPv4:
Subnetting divides an IP address block amongst smaller networks.
Supernetting enables large organizations or network service providers to obtain large blocks of Class C addresses and then to allocate those large blocks to smaller organizations that would otherwise require precious Class B address space.
Private networking allows a corporation to create its own addressing scheme for a private network that does not need to be understood, or routed, by the Internet.
These strategies temporarily have staved off the threat of IP address depletion. Better still, they have encouraged network managers to learn and implement conservative addressing practices.
But they do not offer infinite extensions of current address space. And, they represent substantial administrative and provisioning requirements for small, medium and large enterprises. Not to mention the burden on network service providers, particularly those offering Internet access or Internet-based services.
Simplifying Business
Obviously what's required is a simple and intelligent solution to the problem--a solution that benefits the service provider's business, not one that adds complexity and cost. But if the network needs addresses, it needs addresses, right?
Wrong. We can wean the 800-pound gorilla off its excessive consumption of IP addresses.
To do so requires a sound understanding of the access network. The access network is a crucial domain as far as IP-address consumption goes because, as the part of the network that manages end users and applications, it's the stomach of the beast. But it also holds the potential for significant IP address conservation.
The access network's hunger for addresses is rooted in the typical address allocation scheme, which requires a minimum of four addresses for a single wire: one for the wire (network) address, one for the broadcast address and two for device addresses--for example, the port plus the user PC.
Typically, any port on any routed IAD will be assigned this size of address block. A DSL link to the IAD would additionally require this same size of block. Each additional subscriber adds another four addresses, and so on.
Essentially, then, addressing for a single user device may consume more than four times what might seem to be the logical dietary requirement: one address for one user.
Let's look at the example shown in Figure 1. One customer premises routing device has three subscriber PCs homed off an ISP's network. This totals three wires, plus one wire connecting the customer premises router to the network router, and another wire to the ISP network. That's five wires in total, times four addresses per wire, for a total of 20 addresses to serve three subscribers--nearly seven addresses per subscriber.
Most service providers will be serving tens of thousands of end users, so you can do the math: IP addresses are gobbled up quickly in the access network.
But this is only because the access network until recently has not been tended to. Today, focus is shifting toward the access network as the "last-mile" stretch between service providers and their bandwidth--and service-hungry users.
Truly understanding the access network reveals the need for a next-generation architecture that accomplishes many things, including IP address conservation.
That architecture exists today in the form of a services-aware access architecture, which places key intelligence at the "binding points" of the network.
"Bookending" the access network in this way accomplishes many things:
Secures the access network from end to end;
Grants a service provider greater control over that crucial part of the network it relies on to do business;
Controls traffic across the access network, rather than allowing traffic to enter the core to be managed.
The bookend approach also enables significant reduction in the number of IP addresses required for end users. In fact, the right architecture supports an IP addressing scheme that reduces the dietary requirement of four addresses per line to just one, and in some cases, to a fraction.
Here's how it works.
By binding the network with intelligence at both edges of the access network, a services-aware network architecture has greater control over what goes on between those two points of intelligence. And, those two points speak each other's language.
This enables innovations that control and conserve IP addresses. An innovation like domain switching is a hybrid of IP-level forwarding and label switching that enables "virtual wires" across the access network. Virtual wires extend core services right to the end user's premises.
Because a virtual wire overlays the access network with logical, or virtual access networks (VAN), a subscriber or a subscriber's organization can be assigned its own domain with its own IP address space totally segregated from other domains. The entire domain appears as a single "wire", thus reducing IP address waste.
The total address consumption can be reduced to one address per subscriber device. Here, three subscriber PCs are connected to an intelligent IAD, routed through a multiservice access switch and homed off an ISP network.
Using the "virtual wire" method, the only IP addresses required are, one to identify the ISP network, and one for each subscriber device--the ISP service is essentially extended directly to the subscriber's device. That's about one address per subscriber.
The segregation of data between domains also allows for overlap of address space between domains.
This is an example in which IP addresses are assigned sequentially and contiguously. But a services-aware architecture also allows for noncontiguous IP address assignment, further reducing waste.
In addition, a remote authentication dial-in user service (RADIUS) server dynamically can assign IP addresses to subscribers just for the length of a single session. Once the user session is terminated, that dynamically assigned address is returned to a pool of addresses to await a new subscription request.
If the access network is understood well and tended to with solutions that meet its unique requirements, IP address consumption actually can be reduced to a fraction. And no one will starve.
So, the network can be weaned. IP address consumption can be substantially reduced. And service providers can stop worrying about IP addressing hassles and let a smart, services-aware network do the work.
Terry Skemer was senior network architect at the former Sedona Networks. With more than 20 years of telecommunications network and engineering experience, he holds numerous patents in voice-over-packet and packet-networking technologies.
IPv4 Address Classes | |||
| Address Class | Type(bits reserved for network identifier) | (Maximum number of hosts per) | (Maximum number of networks) |
| A
B C D, E | Eight
Sixteen Twenty-fours Reserved | 16.7 M
65,534 245 1/8th of total address space | 126
16,000+ 2.1 M |
