Active Directory Replication

Replication Behavior

Replication behavior is consistent and predictable; given a set of changes to a specific replica, it can be predicted that changes are going to be propagated to all other replicas.

The following key points are central to understanding the behavior of Active Directory replication:

An object is available for replication as soon as it is written. Writes to single objects are atomic, so "partially written" objects are not possible.
Objects are not necessarily replicated in the order in which they are updated.
After an update cycle is initiated, a specific replication cycle sends all available changes from the source replica to the destination replica, including changes that occur while the replication cycle is in progress.
Replication is store-and-forward and "ripples" through a set of connected replicas.
Multimaster conflict resolution is guaranteed reliable even if clocks become unsynchronized or move backward.
The graph of replication connections is not always a spanning tree (which by definition does not contain redundant links) — the graph can, and generally does, contain cycles. Redundant connections reduce replication latency, especially in case of failure. A propagation-dampening mechanism eliminates redundant replication.
Replication within a site is triggered by a change notification mechanism when an update occurs, moderated by a short, configurable delay (because groups of updates frequently occur together).
Replication between sites typically occurs at scheduled intervals (change notification between sites is optional).
The system is resilient in the face of load spikes and temporary failures.
The replication system is designed to be stable. Every time that a replication destination receives information from a replication source, the destination becomes more up-to-date. Recovery from failures involves a minimum of extra work.
Store and forward replication makes efficient use of WAN links — each update crosses an expensive link only once and is compressed.
Replication topology is managed automatically and optimizes existing connections.

Microsoft® Exchange Server version 5.5 uses sites and replication somewhat differently than Windows 2000 does. For Exchange administrators who are familiar with Exchange directory replication, Table 6.1 provides a summary of the significant differences between Exchange and Active Directory replication. Treating Active Directory as if it were an Exchange directory does not make optimal use of Active Directory.

Table 6.1 Differences Between Windows 2000 and Exchange Server Version 5.5 Directory Services

Windows 2000 Directory Service	Exchange Server 5.5 Directory Service
Master replicas accept updates independently without communicating with other master replicas.	Each directory service object is mastered in a specific site that can be determined by its distinguished name. Updates are multimaster within the master site.
The basis for replication is the object GUID. When an object is renamed, its GUID does not change, so renaming the object cannot lead to replication errors.	The basis for replication is distinguished names. Therefore, to avoid problems, Exchange does not rename objects.
Replicates an update by transmitting only the changed attributes.	Replicates an update by transmitting the entire object.
Supports compression of replication data between sites over remote procedure call (RPC) or Simple Mail Transfer Protocol (SMTP) transports.	Supports compression of replication data between sites over SMTP transport only.
Supports servers that contain only a subset of the objects in the entire directory in addition to Global Catalog servers, which contain all objects but only a partial set of attributes.	Holds a full replica of the directory in each directory server. (The schema in Exchange is site-specific and is not replicated out of its site.)
Has a flexible replication topology (including choice of transports).	Has a replication topology between sites that is limited to a spanning tree, which cannot contain redundant links. Replication transport between sites is limited to e-mail.
Uses sites to help generate replication topology and to help clients perform intelligent replica selection; but sites are not tied to directory partitioning.	Uses sites to generate replication topology, but sites are also the unit of directory partitioning.