Session partitioning gives the administrator the ability
to filter or reduce the number of destinations that the session
object gets sent to by the replication service. You can also configure
session partitioning by specifying the number of replicas on the replication
domain. The single replica option is chosen by default. Since the
number of replicas is global for the entire replication domain, all
the session managers connected to the replication domain use the same
setting.
- Single replica
- You can replicate a session to only one other server, creating
a single replica. When this option is chosen, a session manager picks
another session manager that is connected to the same replication
domain to replicate the HTTP session to during session creation. All
updates to the session are only replicated to that single server.
This option is set at the replication domain level. When this option
is set, every session manager connected to this replication domain
creates a single backup copy of HTTP session state information on
a backup server.
- Entire domain
- Each object is replicated to every application server that is
configured as a consumer of the replication domain. However, in the
peer-to-peer mode, this topology is the most redundant because all
servers replicate to each other, and as you add servers, more processors and
memory are needed to manage
replication. Entire domain is most useful for dynamic caching replication.
Redundancy does not affect the client/server mode because clients replicate
only to servers that are set to server mode.
Attention: Do
not use entire domain in a large topology, as the infrastructure cannot
handle the large number of connections that are necessary to support
this configuration.
- Specific number of replicas
- You can specify a specific number of replicas for any entry that
is created in the replication domain. The number of replicas is the
number of application servers that you want to replicate in the domain.
This option eliminates redundancy that occurs in a full group replica,
and also provides additional backup over a single replica. In peer-to-peer
mode, the number of replicas cannot exceed the total number of application
servers in the cluster. In the client/server mode, the number of replicas
cannot exceed the total number of application servers in the cluster
that are set to server mode.
On multiple application server nodes,
the number of replicas must match the number of application servers
on a node to ensure that a backup exists on a different node. For
example, if you have two nodes with three application servers on each
node, you would want to set the number of replicas to three. Location
of backups is randomly selected and in a worse case scenario, the
application server backups might be selected on the same node. Setting
number of replicas to three ensures that at least one backup exists
on a different node.
Example
Node A has application servers A1, A2, and A3. Node B has
application servers B1, B2, and B3. If A1 selects its backups to be
on A2 and A3, then it is forced to select an application server on
node B because number of replicas is set to three.