System Capacity and Service Group Load

The system attribute Capacity sets a fixed load-handling capacity for servers. Define this attribute based on system requirements. The service group attribute Load sets a fixed demand for service groups. Define this attribute based on application requirements.

When a service group is brought online, its load is subtracted from the system's capacity to determine available capacity, which is maintained in the attribute AvailableCapacity.

When a failover occurs, HAD determines which system has the highest available capacity and starts the service group on that system. During a failover involving multiple service groups, failover decisions are made serially to facilitate a proper load-based choice.

System capacity is a soft restriction; in some situations, value of the Capacity attribute could be less than zero. During some operations, including cascading failures, the value of the AvailableCapacity attribute could be negative.

Static Load versus Dynamic Load

Dynamic load is an integral component of the Service Group Workload Management framework. Typically, HAD sets remaining capacity with the function:

AvailableCapacity = Capacity - (sum of Load values of all online service groups)

If the DynamicLoad attribute is defined, its value overrides the calculated Load values with the function:

AvailableCapacity = Capacity - DynamicLoad

This enables better control of system loading values than estimated service group loading (static load). However, this requires setting up and maintaining a load estimation package outside VCS. It also requires modifying the configuration file main.cf manually.

Note that the DynamicLoad (specified with hasys -load) is subtracted from the Capacity as an integer and not a percentage value. For example, if a system's capacity is 200 and the load estimation package determines the server is 80 percent loaded, it must inform VCS that the DynamicLoad value is 160 (not 80).

Overload Warning

Overload warning provides the notification component of the Load policy. When a server sustains the preset load level (set by the attribute LoadWarningLevel) for a preset time (set by the attribute LoadTimeThreshold), the loadwarning trigger is invoked. For a full description of event management with triggers, see VCS Event Triggers. For details on the attributes cited above, see System Attributes.

The loadwarning trigger is a user-defined script or application designed to carry out specific actions. It is invoked once, when system load exceeds the LoadWarningLevel for the LoadTimeThreshold. It is not invoked again until the LoadTimeCounter, which determines how many seconds system load has been above LoadWarningLevel, is reset.

Limits and Prerequisites

System limits and service group prerequisites strengthen the Load policy.

Limits is a system attribute and designates which resources are available on a system, including shared memory segments and semaphores.

Prerequisites is a service group attribute and helps manage application requirements. For example, a database may require three shared memory segments and 10 semaphores. VCS Load policy determines which systems meet the application criteria and then selects the least-loaded system.

If the prerequisites defined for a service group are not met on a system, the service group cannot be brought online on the system.

When configuring these attributes, define the service group's prerequisites first, then the corresponding system limits. Each system can have a different limit and there is no cap on the number of group prerequisites and system limits. Service group prerequisites and system limits can appear in any order.

You can also use these attributes to configure the cluster as N-to-1 or N-to-N. For example, to ensure that only one service group can be online on a system at a time, add the following entries to the definition of each group and system:

System limits and group prerequisites work independently of FailOverPolicy. Prerequisites determine the eligible systems on which a service group can be started. When a list of systems is created, HAD then follows the configured FailOverPolicy.

Using Capacity and Limits

When selecting a node as a failover target, VCS selects the system that meets the service group's prerequisites and has the highest available capacity. If multiple systems meet the prerequisites and have the same available capacity, VCS selects the system appearing lexically first in the SystemList.

Systems having an available capacity of less than the percentage set by the LoadWarningLevel attribute, and those remaining at that load for longer than the time specified by the LoadTimeThreshold attribute invoke the loadwarning trigger.