Replicas serve two purposes.
Providing Fault Tolerance |
 |
If your network covers a large geographical distance, you
might consider placing
partition replicas on a server in another area. This accomplishes
two things.
It allows users in that area to access
your partition more rapidly.
It provides a backup of your partition if a disaster
destroys some local servers and replicas.
You should have enough replicas of every Directory partition
to provide sufficient database backup. If you lose a partition and
do not have a replica of that partition, you could permanently lose
access to a part of your Directory tree.
Directory replication does not provide fault tolerance for
the file system. Only Directory information about objects is replicated.
To provide fault tolerance for your files, you must take advantage
of the host system's fault tolerance features.
Decreasing WAN Link Traffic |
|
If users are accessing the Directory tree through a WAN link,
you can place a read-only replica of the necessary
partition on a local server so they don't need to cross
the WAN link.
Storing a read-only or writable replica on servers that are
across a WAN link can be helpful because it cuts down on the traffic
that has to cross the link when users try to access that partition's
information. Nevertheless, there will be some increase in traffic
due to the synchronization of replicas.
Read-only replicas do not support user login. Do not create
a read-only replica of a partition that users must attach to before
they authenticate to the network.
With a replica of a distant partition stored locally, users
have immediate access to the objects they need. The only time Directory
information crosses the link is when replicas are being updated.
However, remember that every server that carries a replica
must receive all changes to any object within that partition. The
more replicas of a given partition you have, the more time needed
and the more WAN traffic that exists to fully synchronize the replicas.
Before you begin distributing replicas, think about how much
data you want in a partition. Because replicas are stored on servers,
unnecessary information in a replica is an inefficient use of disk
space and network traffic.
If a partition becomes very large, and you only need to replicate
a portion of it, you can use utilities to split the partition and
replicate only the necessary portion.
The following figure shows one way to distribute replicas
across the WAN on our example tree.
This example reflects the following:
Master replicas are stored at each
local site. That is, a server at the New York site stores the master
replica of the [Root] partition, a server at the Los Angeles site
stores the master replica of the MFG partition, etc.
Servers at the Detroit location store replicas of
the [Root] partition and the MFG partition so that information is
locally accessible to users in the Engineering and Quality Assurance
departments.
A server at the Los Angeles office stores a replica
of the Detroit partition so that developers in Los Angeles do not
have to use a WAN link to access information from their counterparts
in Detroit.
A server in the New York office stores a replica
of the Detroit partition to allow local access.
Objects in the Detroit office often access objects
in other parts of the Directory tree across the WAN. (This office
operates as a self-contained business unit within the organization.)
This is only one example of how to place replicas. You must
decide how to best eliminate single points of failure and provide
your users with easy access to information according to your physical
network layout..
Printable version
