Mathew Lodge <firstname.lastname@example.org>
Amanda Jobbins <email@example.com>
5599 San Felipe, 4th Floor
Houston, Texas 77056, USA
Previous INET developing-country workshops and a lot of hard work have enabled many developing countries to make a connection to the global Internet. This paper focuses on the next big challenge facing such countries: how to extend access to a wide range of individuals within that country or region--i.e., the launching of an Internet service.
This paper defines access and service and differentiates the latter from the former. Critical issues for service providers are covered, including:
In the course of their work, the authors have discovered that many would-be Internet service providers (ISPs) need guidance on bridging the gap between gaining Internet access and offering an Internet service. The motivation for writing this paper was to document the answers to those questions that the authors face most frequently in the hope of flattening the learning curve for would-be ISPs. It also makes our lives as consultants easier!
Though this paper is aimed squarely at a commercial audience, we think it is still relevant to educational and nonprofit ISPs. Such organizations face exactly the same challenges as commercial ISPs; the only difference is in the approach to tariffing and billing. Even then, it is typical that the sponsor(s) or subsidizer(s) of educational/free services still want to know the kind of information that normally appears on a bill, even if no money changes hands.
Internet access and Internet service are often confused, as the difference between each is subtle but important. Internet access refers to getting a "pipe to the global Internet" for a single organization. Typically access consists of a dedicated or "leased" line and an agreement with an ISP for them to route traffic to and from your organization. All the traffic destined for the global Internet is generated within your organization, and all traffic received from the global Internet is destined for your organization.
An Internet service, on the other hand, is the situation whereby access can be resold to other organizations (third parties). In this case, the organization is acting as a transit between the global Internet and the third parties. Traffic received from the global Internet may be destined for any of the third parties or for the organization itself. Similarly, traffic from a third party may be destined for one of the other third parties, the global Internet, or the organization offering service.
Hand in hand with offering a service are issues of service management, all of which essentially boil down to one theme: ensuring customer satisfaction. Customer satisfaction protects and grows the revenue of the ISP.
We have labored the point of the difference between access and service for two main reasons:
Something about the Internet phenomenon has made us believe that it is just a "good" business to get into. There are plenty of users, more free PR coverage than one could ever have thought possible, and lots of interesting content to keep the customers coming.
But amongst all this euphoria, the foundation of a successful business endeavor must not be forgotten--the marketing plan!
It is critical that your Internet endeavors are preceded by a thorough andwell-thought-out marketing plan. Put simply, the marketing plan is going to lay out what you are trying to achieve by entering the ISP market, i.e., your financial or strategic goals, what products you are going to sell, to whom, at what price, and how. It is the roadmap for your ISP business.
Underlying the marketing planning process is the production of thorough user forecasts. These forecasts should predict--as far as possible--the number of customers you expect to gain for your service over a two-year period. To attain this figure, you must first have an understanding of the total market for Internet access services segmented into:
Figures for the total potential market will be influenced by variables such as the total population (of adults/companies, etc.), the penetration of PCs and modems, numbers of customers your competitors are known to have, and so on. If you are introducing the first Internet service in your market and are having difficulty predicting the rate of service uptake, you may wish to take a look at the growth curves of other very popular technologies in the consumer market, such as mobile phones.
Predicting your business user base should prove somewhat more straightforward. Interviews should be held with potential major customers to learn first-hand what they want.
Being able to fairly accurately predict the number of customers your service will attract is vitally important when it comes to building your Internet services platform. The technical issues cannot be handled in isolation--the platform should be built to meet the marketing requirements.
Issues such as how many modems are needed in the platform access server are calculated from a knowledge of how many users the service will have in total. It is also a good idea to think about usage patterns in your market. If local calls were free after 10 p.m. in your part of the world, you would naturally find that the peak load on the system would coincide with the free calling period.
At this point another great marketing tool rears its ugly head: the internal audit. The would-be ISP must consider the strengths and weaknesses of its own organization and how these will impact its ability to offer service. It may be that the would-be ISP's existing infrastructure is not designed to cope with the demands of providing consumer Internet access services. For example, the organization may have no mechanism to distribute the client software you will need to provide to all consumer dial-up customers. Or they may not have enough trained personnel to take support calls from the thousands of dial-up customers you could expect to bring on board.
The best advice we can give is this: If you're new to Internet services, go for the customer base with which you already have some familiarity and build from there. In general leased-line services are simpler to provide to consumers than are dial-up services, the latter being more support- and infrastructure-intensive. If you want to minimize effort, you may wish to wholesale access services to other ISPs. In this relationship the other ISPs buy connectivity to the Internet from you but use this to sell their own services.
Marketing communications ("marcoms") and advertising are fun. You can undertake a wealth of exciting activities to promote your Internet service, but remember not to generate too much interest. No, we really mean that!
You must gauge your marcoms activities in light of the user forecasts you have planned to attain, the capacity of the system, and the service level you wish to provide. If you carefully plan the introduction of your Internet service, you will hopefully be one of the ISPs who will gain a reputation for being professional. This is a fairly rare accolade in the world of Internet service provision.
Internet access is typically offered as a package. This bit of marketing jargon just means that a number of elements are bundled together and are priced as one unit. Definitions of the packages offered feed directly into the order processing and fulfillment part of an ISP's organization. For example, the information required from the user for a dial-up account (e.g., what e-mail username they want) must appear somewhere on an order form and also be included in the documentation supplied to the user.
For dial-up users, the basic package typically consists of some software, a dial-up account, and one or more e-mail addresses. Supplying dial-up access software not only helps users by providing them with what they need to get started, but also helps the ISP. If the software is preconfigured as much as possible, then the ISP can eliminate many of the typical calls to the help desk concerning configuration questions and complaints. We find that the majority of help-desk calls relate to software configuration issues, so it makes sense to offer preconfigured software with good manuals.
Dedicated access packages range from the provision of the line to the ISP alone, to a full package including a managed router and firewall on the customer's premises. Which of these is offered depends largely on the skill set and resources of the ISP's operations staff. It also depends on the credibility the ISP has with business customers, who will be the prime customers for this service. We have seen that businesses will not buy from an ISP they feel lacks the ability to support them.
Order processing may well not be too much of an issue for your organization if you already have an order processing procedure in place for other services. The key element of any order processing system is the ability to turn orders around quickly. This simply means processing the order internally: Enter the user account onto a database, run a credit check, set the user up on the system, produce a welcome pack, and send this to the customer with account number and login details and possibly some software.
For dial-up customers the process will be fairly linear. For leased-line customers or customers where the access line is being put in place by a third party, some consideration will have to be given as to how this interface will be managed. With dedicated types of connections, for example, you may be providing a customer premises router to the customer (on a sale or lease basis). The order process will have to initiate the dispatch of someone from technical sales support/engineering to go to the customer's premises and install and test the router.
The price at which an ISP provides an Internet service coupled with the ability to bill customers is absolutely fundamental to success. Without incoming revenue, the ISP will fail. Pricing (tariffing) is somewhat of a black art, and the intricacies of billing and invoicing have proved to be so complex that it has become an industry in its own right. For this reason we will try to consider simple ISP pricing and billing.
When pricing, the key issue is the price sensitivity of Internet products in your market-place. Are you a monopoly provider? Is the market highly competitive? Naturally you will have a good understanding of the state of play in your market, and this knowledge will form an essential input to the pricing decision. You will also need to review the objective for offering Internet services. If you are one of the few companies with a formal planning process, you will have a stipulated financial return to achieve, and this will help you when as you consider how to price your services.
Aside from revenue, an ISP has a number of options as to how you structure your pricing:
For all types of access services, an additional consideration is the cost of the telephone call/leased circuit/ISDN line. For dial-up customers using the public telephone network (PSTN), you may, as in the United States, have an infrastructure that supports free local calls. Alternatively, consider setting up a toll-free line for your Internet service. This will certainly differentiate your service where local calls are expensive.
For ISDN, the cost of access usually does not include the ISDN line, unless you are the provider of ISDN in your market. This raises another issue: one of the price comparison between competitors. For example, if you bundle the ISDN access line and your competitors do not, you must draw attention to this issue very clearly. The customer may not realize that subtle difference until it's too late!
Leased-line services are still primarily charged at a flat rate based on the bandwidth of the pipe.
Billing and pricing are obviously very tightly connected. However, one issue often forgotten is that it actually costs more money to implement a more intricate pricing policy. Usage-based pricing is an example of this--significant amounts of information must be collected to be able to bill customers based on usage.
For example, a measurement of usage in the defined unit must be taken (such as the number of connected minutes for a particular customer). This must be collected and analyzed for a given period (e.g., a month), and if different rates according to peak/off-peak times are applied, then a billing system must calculate the time at which various units were used and multiply them by the appropriate price. Needless to say, a system that simply charges a flat fee per customer account costs somewhat less to implement!
Billing and pricing issues can only be determined to best effect when a solid understanding of the ISP's objectives for launching the service are known. We have seen that this typically occurs when the service provider has a very good understanding of the competitive situation and the price sensitivity of demand for Internet services.
This is essentially what being an ISP is all about: making sure customers can get on the Internet. The three main types of access will be considered in order: dedicated or leased-line connections, analog dial-up, and digital dial-up (ISDN).
The ISP typically has a number of ports on a router with which to offer dedicated access. The goal is to connect as many customers to these ports as possible, as routers offer a limited port density. There are two popular ways to maximize the usage of router ports: frame relay and multiplexing.
The multiplexing approach uses the fact that telephone companies typically multiplex T1 (1.536 Mbps) or E1 (2.048 Mbps) circuits into multiple 64k chunks, because a standard voice circuit consumes 64k of bandwidth. Thus, there is usually a local carrier that can provide multiple 64k circuits multiplexed into a single T1 or E1.
Each 64k circuit can physically terminate at a different customer's premises, but they are all multiplexed together and are provided to the ISP as a single T1 or E1. You can get 24 64k circuits into a T1 and 30 into an E1. Most router vendors offer multiplexed T1/E1 cards that can dramatically improve the port density and lower the port cost. Even if the local carrier cannot provide multiplexed T1/E1 circuits, it may be worthwhile for the ISP to purchase a small multiplexer because of the large cost benefits of this approach.
The frame relay approach allows multiple "virtual circuits" to be connected to a single physical port. This is more flexible than the multiplexing approach as the virtual circuits are not constrained to 64k--they can be any bandwidth from 64k up to T1/E1. A typical router port offers a full T1 or E1 bandwidth, which can be shared out as multiple frame relay circuits.
There is a cost impact to using frame relay, in that some of the bandwidth is consumed by signaling and transmission overhead. A good rule of thumb is that frame relay overhead consumes 25 percent of the line bandwidth. For a T1 line, this leaves 1.152 Mbps of bandwidth free. You can divide up this remaining bandwidth any way you like--for example, two 512k circuits and a 128k circuit, or nine 128k circuits.
The router's software must be able to handle the frame relay protocol, obviously, but the biggest barrier to this technique is the availability of a public frame relay network. Although there are many carriers offering frame relay in the developed world, it is still scarce in the developing world. Finally, if a customer purchases a full T1 or E1 line, running frame relay over it just consumes bandwidth for transmission overhead with no gain: you only get a benefit when you use frame relay to share the line.
Analog dial-up is justly popular as a low-cost method of connecting to the Internet. However, from a commercial perspective, dial-up access has a lower potential margin than dedicated access. This is because dial-up cost of service is high, while the revenue per customer is relatively low.
Service cost is high because the ISP's modems and access servers (also known as communications servers or terminal servers) are expensive, and dial-up billing is complicated.
The ISP requires modems that are a cut above the bargain $99 variety. The modem must interoperate with a large number of different modems from many vendors and tolerate a large range of line conditions, all while offering the best performance possible. With a large number of modems, the ability to configure and manage them all from a single point becomes important.
All of these features make carrier grade modems expensive. Terminal servers (such as the Cisco 2511 or Livingston Portmaster) are cheap in comparison. Even so, we have seen that the dial-up access equipment can account for between 25 percent and 50 percent of the equipment cost of an ISP system.
Billing is expensive primarily because dial-up users expect to be billed for connection time. The authors recommend billing by connect time because it provides back-pressure on users and encourages them to make more efficient use of the ISPs resources. In situations where the ISP is offering flat-rate billing (i.e., a simple monthly charge with as much connect time as the user likes), we have found that typical user behavior is to log on and then stay on for long periods of time.
This is bad news for the ISP and users alike, who are contending for the limited number of modems on offer, and also the bandwidth to the global Internet. With flat-rate billing, there is no incentive for the user to log off. With connect-time billing, the user is aware that the meter is running, and we have seen that a typical user will log on, do what they need to do, and log off. What kind of billing you decide to offer is also influenced by local market conditions (see section 5).
There are typically two main uses for ISDN dial-up: as a fast "pipe" for individual users, and as a low-budget LAN Internet connection. The former is just like analog dial-up, but with a faster (digital) line. The latter is just like a dedicated (leased-line) connection, but with a line that isn't dedicated. This is also known as "dial on demand," because the line is brought up only when there are data to be transmitted.
The reason for differentiating between these two types of service mainly has to do with routing issues. Dial-on-demand customers need the same type of interdomain or exterior routing that dedicated customers need (see section 7). Individual ISDN users can use the same simplified routing that is used for analog dial-up.
The ISP must also be careful to price the service at a level that makes sense for the customer. Dial-on-demand ISDN has to work out cheaper than dedicated 64k connectivity, including any call charges levied by the carrier. Thus, you should calculate at which point the 64k dedicated line becomes more attractive (e.g., ISDN is cheapest until the line is in use for more than 40 hours per week). Your customers will certainly do this calculation as well!
An ISP needs two sets of IP address space: IP addresses for the ISP's own infrastructure, and IP addresses to lease to customers. We will explain why we're so precise as to say "lease" in a moment, but first an overview of how IP address allocation works.
The Internet Assigned Numbers Authority (IANA) owns the IP address space. "IP address space" means the set of all IP addresses. Visualize it as a grid of boxes, with each box representing an IP address. The IANA has delegated the distribution of the IP address space to regional registries: the InterNIC, RIPE, and APNIC. Previous IP address allocation strategies mean that there is little IP address space left, and conserving this space is therefore a number one priority with the registries. Note that only IANA owns the address space. Everyone else (this means you) is leased addresses.
The current IP address allocation strategy can be summed up as provider-based addressing. The basic rules are pretty simple:
The main problem with these rules is that they imply that renumbering is necessary. For example, say you get some IP addresses from provider A and use them to number your networks. All is well until you decide to change providers: maybe provider A is giving you bad service or charging much more than provider B. At this point, you must hand back those IP addresses to provider A, get some new addresses from provider B, and renumber your network.
The problem with this is that renumbering technology is pretty raw and underdeveloped, and lots of current equipment relies on statically assigned IP addresses. This makes renumbering a royal pain, and a real revenue loser: your system is down until you get it renumbered and have ironed out all the renumbering bugs. Unfortunately, no one has a practical solution for this dilemma--yet . But why can't you just take those addresses from provider A and assign them to provider B?
Provider-based addresses cannot be transferred because of a thing called CIDR: Classless Inter-Domain Routing. CIDR was born out of another important issue facing the global Internet: The full Internet routing table grows at a rate of around 100 percent per year. Backbone (or "top level") ISPs must carry a routing table that contains routes for all the networks in the Internet. This is a consequence of the way that Internet routing works and the way that commercial routers have evolved.
This trend is a problem because the number of routes in the table will soon outstrip the amount of memory available in backbone routers. Building bigger routers is not a solution because routers would have to become exponentially large to solve the problem.
Simplifying considerably, CIDR approaches this problem by allowing a single routing table entry to provide routing information for many Internet networks. In CIDR, one routing table entry can provide information for one network, or two networks, or four networks, or any power of two networks. This is also known as route aggregation, because a number of routing table entries make up (or aggregate) into a single entry. An important prerequisite of this approach is that the networks must be consecutively numbered.
So what does this have to do with provider-based addressing? Providers are given "CIDR blocks"--consecutively numbered networks that can be aggregated using CIDR. This is why you must hand provider-based addresses back to the provider. If you take them elsewhere, the CIDR block is broken into, at best, two pieces (though probably more), which (at best) doubles the number of routing entries required to properly route those networks in the global Internet.
We have described two conflicting goals here: IP address registries want to conserve address space by handing out only as much as is actually required. This means lots of small chunks of address space. Global ISPs want fewer routes in the global routing table. This means fewer, larger chunks of address space.
The bad news is that a prospective ISP must bear both in mind when attempting to get address space. If you are happy with the concept and drawbacks of provider-based addressing, by all means get IP addresses from your provider. This will usually be quick, at the expense of renumbering pain if you decide to change providers later on. Otherwise, the ISP must make an application to a regional IP address registry.
The Asia-Pacific Network Information Centre (APNIC) handles allocation for countries in the Pacific and in Asia (oddly enough). RIPE (Reseaux IP Européen)  handles Europe and the Middle East. The InterNIC  handles North and South America and everywhere else not covered by RIPE or APNIC. Perhaps surprisingly, at the time of writing, each registry had a different policy for allocating IP addresses. Search their Web and FTP sites for all relevant information and ask for an ISP-specific IP address application form.
The key to a good IP address application is to provide as much relevant information as possible about why you need the addresses and to show you understand the issues described above. Making a realistic application also helps: A start-up ISP is extremely unlikely to be able to justify getting a class B network. Two to four class Cs, as required, is a good starting point. We recommend that you get assistance from someone who has made successful applications for ISPs in the past.
Most of the registries have a "slow start" policy for ISPs, which means that you will get a small number of networks (perhaps a pair of class Cs) to begin with. More are then assigned as the ISP's business grows and the ISP can show how the address space has been utilized. The IP registries get hundreds of applications from people wanting to become ISPs who have practically no knowledge of what it involves, all requesting big blocks of IP address space because their business plan shows they're going to be the next UUnet. The IP registries just don't have the time or resources to evaluate and educate every would-be ISP, and slow start policies are the result.
Finally, make sure you pay any fees due the IP registry. This may sound like rather obvious advice, but we have had to help ISPs who have missed important business deadlines simply because they did not pay their dues.
Because an ISP is providing a transit service, routing is more complicated than for simple access by a single organization. The ISP and each dedicated customer should have an autonomous system (AS) number--these are allocated by the InterNIC. Border or exterior routing defines how data pass between autonomous systems.
It is important that dedicated (leased-line) customers have their own AS and that their networks are not part of the ISP's AS. This is because the AS defines an administrative boundary between organizations and limits the flow of routing information from one AS to another. If a customer is inside the ISP's AS, it is much easier for errors in the configuration of the customer's network (usually routing errors) to propagate into the ISP's service equipment. Configuration errors mean down-time, angry customers and lost revenue for the ISP.
The only well-supported border routing protocol is called BGP-4 (for Border Gateway Protocol version 4); this is used to exchange information between the ISP and each dedicated customer and between the ISP and its upstream provider.
The authors have found that, after IP addresses, domain names are the second largest area of misunderstanding and concern for would-be ISPs. We have also found that this confusion is encouraged and leveraged by unscrupulous practitioners. So we would like to start out by clearly stating that domain names, IP addresses, and the Domain Name System (DNS) are wholly separate entities. Where you got your IP addresses has no bearing on what domain name you use, and neither affects where your DNS service is operated.
Simply stated, the DNS system provides a mapping between domain names and IP addresses, and also between IP addresses and domain names. Domain names themselves form a hierarchy. Once again, IANA is the overall authority for domain names.
When the Internet was small and U.S.-centric, it made sense to divide domain names up into categories. These were known as Top Level Domains (TLDs), and they are the familiar .com, .net, .gov, .edu, and .mil. As the Internet grew, each country got its own TLD. The IANA selected the ISO two-letter digraph system to name the country domains, so that France got the TLD .fr, Japan got .jp, and so on. Using the digraphs offered the twin benefits of using an international standard and of removing the IANA from disputes about what constitutes a country.
Perhaps regrettably, the U.S.-centric TLDs remained, so that there were the country-specific domains as well as the older .com, .net, etc., TLDs. This raises the question of the domain in which an organization should register itself: a country-specific domain or .com/.net.
Many organizations want to register in the .com domain because their managements perceive that having a domain name ending in .com is to be more closely identified with the Internet than any other domain name. Regardless of whether this is actually the case, it has led to the oversubscription of .com. One reason for the trend toward longer .com domain names is that at the beginning of 1996, over 90 percent of the domain names of eight characters or less had been registered. Perhaps the canonical example of how ridiculous this becomes is the domain name batmanforever.com. The whole problem is exacerbated by the fact that U.S.-based organizations have no incentive whatsoever to register in the .us domain.
The InterNIC handles the registration of all top-level domains (including country domains), so if you are the first ISP for a given country, you need to contact the InterNIC. You can check who operates a TLD (if anyone) by performing a WHOIS look-up at rs.internic.net. The InterNIC also handles subdomain registrations in .com and .net.
Registration within a subdomain of a country is handled by the delegated authority for that country. This is usually (but not always) the organization that operates the DNS server(s) for the top-level country domain. Thus to register a commercial organization in Japan (say compusurf.co.jp), you'd contact the organization listed in the InterNIC WHOIS database who runs the .jp domain.
It is usually the case that you must have an operational DNS system before you can be allocated a domain name. For would-be ISPs this can be a problem, as they often don't have an operational DNS system when they need to make the application for the domain name (it's hard to prepare brochures, manuals, and other marketing materials with long lead times when you aren't certain of the domain name). In this case, ask your provider or another willing organization if they will configure their DNS system to operate the domain (this is sometimes known as proxy DNS, but there is no actual proxying in the same sense as WWW or FTP proxies). When you get your own DNS system up and running, you can take over the primary DNS function.
DNS itself is just a large, global distributed database. We are not going to explain the operation of DNS in any detail, as that has already been done extremely well in a number of places, but particularly . We are going to mention primary and secondary DNS because this is something else we find is misunderstood.
A primary DNS server is the authority on a particular set of mappings from domain names to IP addresses and vice versa. A secondary DNS server is a fall-back server that receives updates from the primary DNS server so that it can answer queries. There can be many secondary servers for a primary server.
It is advisable to have another secondary server without a common point of failure. Power supplies and physical networks are common points of failure, so it helps if you can locate the secondary somewhere completely unrelated to your site.
You can ask your provider, but check to see if they have any point of failure that is common to the both of you (e.g., if your provider has a single link to a further upstream provider, that is a common point of failure; if the link goes down, both you and your provider's DNS systems are disconnected from the global Internet).
Reciprocal secondary DNS arrangements are common. For example, a university may provide secondary DNS services for you if you, in return, provide secondary DNS for the university. Make sure that any organization with whom you are seeking this kind of arrangement has enough resources and connectivity to be suitable as a fall-back for yourself.
This paper has attempted to straddle the line between business and technology by considering issues on both sides. This is the very essence of a successful ISP: Both commercial and technical issues must be carefully considered, combined, and executed. It is for this reason that the strongest ISPs are forged from the strongest teams: good marketing people and good technical people.
There are many other issues that the authors could not cover due to space restrictions. Hopefully this paper has given any prospective ISPs a good grounding, but there is no substitute for experience: No amount of advice can really convey how exciting launching an Internet service can be.
 The PIER working group of the IETF (Practical Internet / Enterprise Renumbering). http://www.ietf.cnri.reston.va.us/html.charters/pier-charter.html
 APNIC. http://www.apnic.net/
 RIPE. http://www.ripe.net/
 InterNIC, operated by AT&T and Network Solutions Inc. http://rs.internic.net/
 Paul Albitz, Cricket Liu, DNS and BIND, O'Reilly and Associates, ISBN: 1-56592-010-4, 1992.