| eDirectory uses a slow-but-sure convergence algorithm to replicate changes from a replica server to its peers in a replication ring. A replica server can manage only a single Directory Information Base (DIB), but a DIB may contain replicas of multiple partitions. Replication uses a batch update mechanism. The period for which changes are accumulated in a replica server is adjustable, from only a few seconds to a few hours, but it defaults to 30 minutes for NDS 6 and 60 minutes for NDS 7 and higher; this is known as the "heartbeat" interval. Changes to attributes (such as password ) that are flagged Sync Immediate will be scheduled for immediate synchronization. A background thread that would yield or postpone its operation if a request for a Create, Modify, or Delete operation were received handles synchronization operations. This causes a delay, or latency, in replication. Fortunately, in many instances, partitioning can minimize this delay. You can partition a tree such that update operations are spread across multiple partitions. Placing these volatile partitions on different servers helps to minimize the peak update load because the partitions are now distributed. For example, if a tree has three containers that are volatile (whose subordinate objects undergo modifications frequently), you should isolate each container into a partition and place the partitions on separate servers. The larger the peak update rate, the smaller the replica ring; but bear in mind that a ring should be designed with at least two servers (three is recommended) for fault tolerance reasons. If the entire server farm is front-ended by a load-balancing switch, you should configure the switch to direct all requests to the primary servers (ones holding the Master replicas) and fail over to the secondary. NOTE The By Server synchronization method (discussed in the "Multithreaded Synchronization" section in Chapter 6, "Understanding Common eDirectory Processes") may also help reduce the replication latency by outbounding multiple partitions to multiple unique servers at one time.
| 