It's very easy to get bogged down in the early stages of the namespace design without actually progressing much further. The stumbling block seems to be that it feels conceptually wrong to have only one domain, yet administrators can't put their finger on what the problem is. Experienced Windows NT administrators who manage multiple domains seem to find this much more of a problem than those coming from another operating system.
If you follow the guidelines in the initial steps of the namespace design, you quite probably will end up with one domain to start with. That's the whole point of the design process: to reduce the number of domains you need. Yet NT administrators tend to feel that they have conceptually lost something very important; with only one domain, somehow this design doesn't "feel right."
This is partly a conceptual problem: a set of domains with individual objects managed by different teams can feel more secure and complete than a set of Organizational Units in a single domain containing individual objects managed by different teams. It's also partly an organizational problem and, possibly, a political problem. Putting in an Active Directory environment is a significant undertaking for an organization and shouldn't be taken lightly. This change is likely to impact everyone across the company, assuming you're deploying across the enterprise. Changes at that level are likely to require ratification by a person or group who may not be directly involved on a day-to-day basis with the team proposing the change. So you have to present a business case that explains the benefits of moving to Active Directory.
Following our advice in this chapter and Microsoft's official guidelines from the white papers or Resource Kit will lead most companies to a single domain for their namespace design. It is your network, and you can do what you want. More domains give you better control over replication traffic but may mean more expense in terms of hardware. If you do decide to have multiple domains but have users in certain locations that need to log on to more than one domain, you need DCs for each domain that the users need in that location. This can be expensive. We'll come back to this again later, but let's start by considering the number of domains you need.
If the algorithm we use to help you determine the number of domains gives you too small a figure in your opinion, here's how you can raise it:
Have one domain for every single-master and multimaster Windows NT domain that you have. If you are using the Windows NT multimaster domain model, consider the entire set of multimasters as one domain under Active Directory (use Organizational Units for your resource domains).
Have one domain per geographical region, such as Asia-Pacific, Africa, Europe, and so on.
Have extra domains whenever putting data into one domain would deny you the control over replication that you would like if you used Organizational Units instead. It's all very well for us to say that Organizational Units are better, but that isn't true in all situations. If you work through the algorithm and come up with a single domain holding five Organizational Units, but you don't want any of the replication traffic from any of those Organizational Units to go around to certain parts of your network, you need to consider separate domains.
Even Microsoft didn't end up with one domain. They did manage to collapse a lot of Windows NT domains, though, and that's what you should be aiming for if you have multiple Windows NT domains.
There are two parts to this: how you construct a business case itself for such a wide-reaching change and how you can show that you're aiming to save money with this new plan.
Simply stated, your business case should answer two main questions:
Why should you not stay where you are now?
Why should you move to Active Directory?
If you can sensibly answer these two questions, you've probably solved half your business case; the other half is cost. Here we're talking about actual money. Will using Active Directory provide you with a tangible business cost reduction? Will it reduce your Total Cost of Ownership (TCO)? It sure will, but only if you design it correctly. Design it the wrong way, and you'll increase costs.
Imagine first that you have a company with two sites, Paris and Leicester, separated by a 64 Kb WAN link. Now imagine you have one domain run by Leicester. You do not have to place a DC in Paris if it is acceptable that when a user logs on, the WAN link uses bandwidth for items like these:
Roaming user profiles
Access to resources, such as server-based home directories
Application deployment via Microsoft Installer (MSI) files
If authentication across the link from Paris would represent a reasonable amount of traffic, but you do not want profiles and resources coming across the slow link, you could combat that by putting a member server in Paris that could service those resources. You could even redirect application deployment mount points to the local member server in Paris (note that I'm saying member server and not DC here). However, if GPOs themselves won't go across the link, you need to consider a DC in Paris holding all the local resources. That gives you two sites, one domain, and two DCs.
Now let's expand this to imagine that you have a company with 50 WAN locations; they could be shops, banks, suppliers, or whatever. These are the Active Directory sites. Next, imagine that the same company has 10 major business units: Finance, Marketing, Sales, IS, and so on. You really have 3 choices when designing Active Directory for this environment:
Assuming everything else is equal, create a single domain with a DC in whichever sites require faster access than they would get across any link. Now make the business units Organizational Units under the single domain.
Everything is in one domain.
You need as many DCs as you have sites with links that you consider too slow. If you want to count a rough minimum, make it 1 DC per site with more DCs for larger sites; that is a rough minimum of 50 DCs. This is a low-cost solution.
With one forest and one domain, any user can log on quickly anywhere because authentication is always to a local DC.
Every part of the domain is replicated to every other part of the domain, so you have no granularity if you don't want objects from one business unit replicating to DCs everywhere.
Create multiple domains representing the 10 major business units. Place DCs for each business unit in whichever sites require faster access than they would get across any link.
This means more domains than the previous solution, but replication can now be better controlled on a per-business unit basis between sites.
Active Directory cannot host multiple domains on a single DC. This can make for an extremely high cost due to the large number of DCs that you may need. If you need to be able to log on to each of the 10 business unit domains from every site, you need 10 DCs per site, which makes 500 DCs. That's a much more costly solution.
With one forest and multiple domains, any user can log on quickly at any site that has a local DC for her domain; otherwise, she would have to span a WAN link to authenticate her logon and send down her data.
Create multiple domains representing geographical regions that encompass the 50 sites. Make these geographical regions the domains and have each domain hold Organizational Units representing business units that contain only the users from that region.
Even if you end up with 10 geographic regions, the DCs for each region are placed only in the sites belonging to that region. So if there were 5 sites per region (to make the math simple), each of the 5 needs only 1 DC. As the namespace model is a geographic model, you need to place a DC for Europe in the Asia-Pacific region only if the Asia-Pacific region ever has visiting users from Europe who need to authenticate faster than they would across the WAN link from Asia-Pacific to Europe. So the number of DCs that you need is going to be smaller.
Domain replication traffic occurs now only within a region and between regions that has DCs hosting the same domain.
You end up duplicating the business units in all the domains... or maybe not, if some don't need all business unitsyou get the idea.
With one forest and multiple domains, any user can log on quickly at any site that has a local DC for his domain; otherwise he would have to span a WAN link to authenticate his logon and send down his data.
We hope this illustrates that while it is easy to map a simple and elegant design on paper, there can be limitations on the feasibility of the design based on replication issues, DC placement, and cost.
Arguably, there are a number of "best" ways to design depending on whom you talk to. We propose an iterative approach with Active Directory, and this is probably going to happen anyway due to the nature of the many competing factors that come into play. On your first pass through this chapter, you'll get a draft design in hand for the namespace. In Chapter 9, you'll get a draft site and replication design. Then you'll come up against the issue that your namespace design may need changing based on the new draft sites and replication design, specifically on the issues of domain replication and server placement that we have just covered. After you've revised the namespace design, you can sit down and look at the GPO design (using Chapter 7 and Chapter 10) in a broad sense, as this will have an impact on the Organizational Unit structure that you have previously drafted in your namespace design. And so it goes.
While this is the way to design, you will come up against parts of your organization that do not fit in with the design that you're making. The point is to realize that your job is to identify a very good solution for your organization and then decide how to adapt that solution to the real world that your company lives in. One domain may be ideal but may not be practicable in terms of cost or human resources. You have to go through stages of modifying the design to a compromise solution that you're happy with.