How Dynamic Resource Allocation Works
This section explains the process and the factors that poold uses to dynamically allocate resources.
About Available Resources
Available resources are considered to be all of the resources that are available for use within the scope of the poold process. The scope of control is at most a single Solaris instance.
On a system that has zones enabled, the scope of an executing instance of poold is limited to the global zone.
Determining Available Resources
Resource pools encompass all of the system resources that are available for consumption by applications.
For a single executing Solaris instance, a resource of a single type, such as a CPU, must be allocated to a single partition. There can be one or more partitions for each type of resource. Each partition contains a unique set of resources.
For example, a machine with four CPUs and two processor sets can have the following setup:
pset 0: 0 1
pset 1: 2 3
where 0, 1, 2 and 3 after the colon represent CPU IDs. Note that the two processor sets account for all four CPUs.
The same machine cannot have the following setup:
pset 0: 0 1
pset 1: 1 2 3
It cannot have this setup because CPU 1 can appear in only one pset at a time.
Resources cannot be accessed from any partition other than the partition to which they belong.
To discover the available resources, poold interrogates the active pools configuration to find partitions. All resources within all partitions are summed to determine the total amount of available resources for each type of resource that is controlled.
This quantity of resources is the basic figure that poold uses in its operations. However, there are constraints upon this figure that limit the flexibility that poold has to make allocations. For information about available constraints, see Configuration Constraints.
Identifying a Resource Shortage
The control scope for poold is defined as the set of available resources for which poold has primary responsibility for effective partitioning and management. However, other mechanisms that are allowed to manipulate resources within this control scope can still affect a configuration. If a partition should move out of control while poold is active, poold tries to restore control through the judicious manipulation of available resources. If poold cannot locate additional resources within its scope, then the daemon logs information about the resource shortage.
Determining Resource Utilization
poold typically spends the greatest amount of time observing the usage of the resources within its scope of control. This monitoring is performed to verify that workload-dependent objectives are being met.
For example, for processor sets, all measurements are made across all of the processors in a set. The resource utilization shows the proportion of time that the resource is in use over the sample interval. Resource utilization is displayed as a percentage from 0 to 100.
Identifying Control Violations
The directives described in Configuration Constraints and Objectives are used to detect the approaching failure of a system to meet its objectives. These objectives are directly related to workload.
A partition that is not meeting user-configured objectives is a control violation. The two types of control violations are synchronous and asynchronous.
The following events cause asynchronous objective violations:
The contributions of objectives that are not related to workload are assumed to remain constant between evaluations of the objective function. Objectives that are not related to workload are only reassessed when a reevaluation is triggered through one of the asynchronous violations.
Determining Appropriate Remedial Action
When the resource controller determines that a resource consumer is short of resources, the initial response is that increasing the resources will improve performance.
Alternative configurations that meet the objectives specified in the configuration for the scope of control are examined and evaluated.
This process is refined over time as the results of shifting resources are monitored and each resource partition is evaluated for responsiveness. The decision history is consulted to eliminate reconfigurations that did not show improvements in attaining the objective function in the past. Other information, such as process names and quantities, are used to further evaluate the relevance of the historical data.
If the daemon cannot take corrective action, the condition is logged. For more information, see poold Logging Information.