|Oracle® Database Administrator's Guide
11g Release 2 (11.2)
|PDF · Mobi · ePub|
Many operating systems provide tools for resource management. These tools often contain "workload manager" or "resource manager" in their names, and are intended to allow multiple applications to share the resources of a single server, using an administrator-defined policy. Examples are Hewlett Packard's Process Resource Manager or Solaris Containers, Zones, and Resource Pools.
If you choose to use operating-system resource control with Oracle Database, then you must use it judiciously, according to the following guidelines:
If you have multiple instances on a node, and you want to distribute resources among them, then each instance should be assigned to a dedicated operating-system resource manager group or managed entity. To run multiple instances in the managed entity, use instance caging to manage how the CPU resources within the managed entity should be distributed among the instances. When Oracle Database Resource Manager is managing CPU resources, it expects a fixed amount of CPU resources for the instance. Without instance caging, it expects the available CPU resources to be equal to the number of CPUs in the managed entity. With instance caging, it expects the available CPU resources to be equal to the value of the
CPU_COUNT initialization parameter. If there are less CPU resources than expected, then the Oracle Database Resource Manager is not as effective at enforcing the resource allocations in the resource plan. See "Managing Multiple Database Instances on a Single Server" for information about instance caging.
The dedicated entity running all the instance's processes must run at one priority (or resource consumption) level.
The CPU resources assigned to the dedicated entity cannot be changed more frequently than once every few minutes.
If the operating-system resource manager is rapidly changing the CPU resources allocated to an Oracle instance, then the Oracle Database Resource Manager might not manage CPU resources effectively. In particular, if the CPU resources allocated to the Oracle instance changes more frequently than every couple of minutes, then these changes might not be observed by Oracle because it only checks for such changes every couple of minutes. In these cases, Oracle Database Resource Manager can over-schedule processes if it concludes that more CPU resources are available than there actually are, and it can under-schedule processes if it concludes that less CPU resources are available than there actually are. If it over-schedules processes, then the
MAX_UTILIZATION_LIMIT directives might be exceeded, and the CPU directives might not be accurately enforced. If it under-schedules processes, then the Oracle instance might not fully use the server's resources.
Process priority management must not be enabled.
Management of individual database processes at different priority levels (for example, using the
nice command on UNIX platforms) is not supported. Severe consequences, including instance crashes, can result. Similar undesirable results are possible if operating-system resource control is permitted to manage the memory to which an Oracle Database instance is pinned.