Inside Windows Server 1002003 [Electronic resources] نسخه متنی

Designing Site Architectures

Any design involves applying available features with adequate safeguards for failures. When laying out sites for your organization, keep the following guidelines in mind:

• Minimize convergence time without swamping WAN links.




• Localize client access to domain controllers for authentication and Active Directory services. This includes Global Catalog queries.




• Prepare for possible failure of bridgehead servers.




• Prepare for possible loss of WAN links.




• Ensure that all naming contexts have a replication path through each site.

The site plan depends on the network infrastructure. Begin with an evaluation of that infrastructure. Figure 7.8 shows a typical WAN layout for a medium-sized organization. The company has four offices scattered around the Southwest. They could just as easily be offices in the same city or overseas. The issue is not distance but the speed and reliability of the connections.

Here are the key aspects of the infrastructure that affect site design:

• Connections between offices use WAN links of various speeds. Some links, like the ISDN line, are expensive if used continually. This requires costing out the site links to ensure replication takes place over an appropriate path.




• The network links, with the exception of the ISDN line, can handle respectable volumes of traffic. This makes it possible to use short inter-site polling frequencies.




• The forest contains two domains. You must ensure that domain controllers from both domains can obtain updates from partners in the same domain. You must also ensure that Global Catalog servers can obtain updates from other GC servers.




• All locations are well connected, so there is no need to use SMTP or to set special replication schedules.

Figure 7.9 shows a provisional site plan for the infrastructure shown in Figure 7.8.

In the example, each office is assigned a site. You might be tempted to put offices connected by high-speed links in the same site to minimize convergence time, but you have a secondary objective to localize client access. Using separate sites ensures that clients in each office authenticate on local domain controllers. Even if the offices were connected by DS-3 or SONET links, you would want to define separate sites in view of the objective to minimize the impact of a lost link.

The IP Subnet objects use the IP addresses assigned to each office by the routing infrastructure. This permits domain controllers to guide a client to the correct site based on the client's IP address.

The Site Link objects mimic the underlying network infrastructure only to the extent necessary to define the inter-office topology. There is no need to create Site Links for every routing path. Site links do not tell Active Directory how to route individual replication packets. They simply define the underlying connection lines. The primary concern is to tell the ISTG about available paths to the other sites so it can create appropriate Connection objects between the bridgeheads.

The Site Link names include the names of the end-points so that you can tell at a glance where the link goes. This works like rural highway names in the Midwest. For example, in Cincinnati, the Carmel-Tabasco road connects the town of Carmel to the town of Tabasco.

You can also include the connection type in the name to differentiate multiple links between the same sites. For example, the primary connection between the Houston office and the Albuquerque office is a frame relay PVC with a Committed Information Rate (CIR) of 512Kbps, so the name would be Hou_Alb_512.

The link costs reflect the speed of the underlying WAN links. The costs tell the bridgeheads where to pull replication. In the example, the bridgehead in Albuquerque would replicate from the Phoenix bridgehead through the bridgehead in Houston. It uses this path rather than pulling directly from Phoenix using the ISDN line because the cost on the Site Link to Houston is lower than that of Phoenix. In the example, costs are assigned as follows:

• The fastest connection is given a cost of 5. This gives some headroom for the future if a faster link is installed.




• The remaining links costs are in ratio to the fastest link. For instance, the 512Kbps links are 1/3 of a full T-1, so the costs are 3*5 or 15. The cost of the ISDN link is 60 (1/12 of a full T-1, or 12*5).

The polling frequency assigned to each fast link is set to 15 minutes. This is the minimum replication interval and yields the fastest convergence. The polling frequency for the ISDN line is set to 30 minutes to reduce the load on the slow link.

The plan designates preferred bridgehead servers. This is not a required step. The KCC will select bridgeheads automatically. Designating a set of preferred bridgeheads is desirable when you want to ensure that the job is given to the most capable machines. It is important that you clearly label your preferred bridgeheads so that the operations staff does not inadvertently take them out of service and stop replication from a site.

Finally, the plan has to designate the location of domain controllers and Global Catalog servers. Here is a set of guidelines:

• Small offices with a few users are not given a local domain controller. The office is still defined as a site so that clients authenticate to the domain controllers in the closest office.




• Slightly larger sites are given a local domain controller that is not configured as a Global Catalog server. The domain controller is configured to cache Global Catalog information so that a cut WAN link does not prevent users from authenticating.




• Large sites are given both a DC and a GC. This ensures that sufficient domain controllers are available for handling peak authentication requests. It also maintains link continuity between sites containing different domains. In practice, you could make each domain controller a GC if you have fast, reliable servers.

You should configure the sites in Active Directory before deploying controllers in the various sites. This permits the system to place the domain controllers in the correct site as they are promoted. Otherwise, you must move the server to the correct site manually