Database Reference
In-Depth Information
STATISTIC VALUE
------------------ -----
Servers In Use 4
Servers Available 8
Servers Started 46
Servers Shutdown 34
Servers Highwater 12
Servers Cleaned Up 0
In a RAC environment, every database instance that is part of the cluster has its own slave processes pool. When a
SQL statement is executed in parallel, slave processes can be allocated either locally, remotely from a single database
instance, or from several database instances. The following methods are available to control the database instances
which the slave processes are allocated from:
•
The
parallel_instance_group
and
instance_groups
initialization parameters can be used
to restrict the allocation of slave processes to specific database instances. With the
instance_
groups
initialization parameter, you specify which groups each database instance lies in. With
the
parallel_instance_group
initialization parameter, you specify which group the slave
processes are allocated from. Since version 11.1, the
instance_groups
initialization parameter
is deprecated. So the method just described should be used in version 10.2 only.
•
From version 11.1 onward, the allocation of slave processes is service-aware. Slave processes
are allocated only from database instances running the same service that is being run by the
session executing the SQL statement in parallel. As of version 11.1 this method replaces the
previous one.
•
From version 11.2 onward, only local slave processes are allocated when the
parallel_force_
local
initialization parameter is set to
TRUE
(the default is
FALSE
).
Because the configuration of the slave processes pool is database instance specific, in a 12.1 multitenant
environment, it's not possible to set the
parallel_min_servers
and
parallel_max_servers
initialization parameters
at the PDB level. However, it is possible to control the allocation of slave processes through services or the
parallel_
force_local
initialization parameter at the PDB level.
Memory Utilization
The table queues used for communication between processes are memory structures that can be allocated from
either the shared pool or the large pool. However, using the shared pool for them is
not
recommended. The large pool,
which is specialized for nonreusable memory structures, is a much better choice. There are two configurations that
lead to a utilization of the large pool for table queues:
•
Automatic SGA management is enabled through the
sga_target
or
memory_target
initialization parameters (the latter is available as of version 11.1 only).
•
The
parallel_automatic_tuning
initialization parameter is set to
TRUE
. Note that this
initialization parameter is deprecated. However, if you don't want to use automatic SGA
management, setting it is the only way to use the large pool for parallel processing.
■
despite its name, the
parallel_automatic_tuning
initialization parameter does only two simple things. First,
it changes the default value of several initialization parameters related to parallel processing. Second, it instructs the
database engine to use the large pool for the table queues.
Note
