Load Balancer for the PowerCenter Integration Service
The Load Balancer is a component of the PowerCenter Integration Service that dispatches tasks to PowerCenter Integration Service processes running on nodes in a grid. It matches task requirements with resource availability to identify the best PowerCenter Integration Service process to run a task. It can dispatch tasks on a single node or across nodes.
You can configure Load Balancer settings for the domain and for nodes in the domain. The settings you configure for the domain apply to all PowerCenter Integration Services in the domain.
You configure the following settings for the domain to determine how the Load Balancer dispatches tasks:
- •Dispatch mode. The dispatch mode determines how the Load Balancer dispatches tasks. You can configure the Load Balancer to dispatch tasks in a simple round-robin fashion, in a round-robin fashion using node load metrics, or to the node with the most available computing resources.
- •Service level. Service levels establish dispatch priority among tasks that are waiting to be dispatched. You can create different service levels that a workflow developer can assign to workflows.
You configure the following Load Balancer settings for each node:
- •Resources. When the PowerCenter Integration Service runs on a grid, the Load Balancer can compare the resources required by a task with the resources available on each node. The Load Balancer dispatches tasks to nodes that have the required resources. You assign required resources in the task properties. You configure available resources using the Administrator tool or infacmd.
- •CPU profile. In adaptive dispatch mode, the Load Balancer uses the CPU profile to rank the computing throughput of each CPU and bus architecture in a grid. It uses this value to ensure that more powerful nodes get precedence for dispatch.
- •Resource provision thresholds. The Load Balancer checks one or more resource provision thresholds to determine if it can dispatch a task. The Load Balancer checks different thresholds depending on the dispatch mode.
Configuring the Dispatch Mode
The Load Balancer uses the dispatch mode to select a node to run a task. You configure the dispatch mode for the domain. Therefore, all PowerCenter Integration Services in a domain use the same dispatch mode.
When you change the dispatch mode for a domain, you must restart each PowerCenter Integration Service in the domain. The previous dispatch mode remains in effect until you restart the PowerCenter Integration Service.
You configure the dispatch mode in the domain properties.
The Load Balancer uses the following dispatch modes:
- •Round-robin. The Load Balancer dispatches tasks to available nodes in a round-robin fashion. It checks the Maximum Processes threshold on each available node and excludes a node if dispatching a task causes the threshold to be exceeded. This mode is the least compute-intensive and is useful when the load on the grid is even and the tasks to dispatch have similar computing requirements.
- •Metric-based. The Load Balancer evaluates nodes in a round-robin fashion. It checks all resource provision thresholds on each available node and excludes a node if dispatching a task causes the thresholds to be exceeded. The Load Balancer continues to evaluate nodes until it finds a node that can accept the task. This mode prevents overloading nodes when tasks have uneven computing requirements.
- •Adaptive. The Load Balancer ranks nodes according to current CPU availability. It checks all resource provision thresholds on each available node and excludes a node if dispatching a task causes the thresholds to be exceeded. This mode prevents overloading nodes and ensures the best performance on a grid that is not heavily loaded.
The following table compares the differences among dispatch modes:
Dispatch Mode | Checks resource provision thresholds? | Uses task statistics? | Uses CPU profile? | Allows bypass in dispatch queue? |
|---|
Round-Robin | Checks maximum processes. | No | No | No |
Metric-Based | Checks all thresholds. | Yes | No | No |
Adaptive | Checks all thresholds. | Yes | Yes | Yes |
Round-Robin Dispatch Mode
In round-robin dispatch mode, the Load Balancer dispatches tasks to nodes in a round-robin fashion. The Load Balancer checks the Maximum Processes resource provision threshold on the first available node. It dispatches the task to this node if dispatching the task does not cause this threshold to be exceeded. If dispatching the task causes this threshold to be exceeded, the Load Balancer evaluates the next node. It continues to evaluate nodes until it finds a node that can accept the task.
The Load Balancer dispatches tasks for execution in the order the Workflow Manager or scheduler submits them. The Load Balancer does not bypass any task in the dispatch queue. Therefore, if a resource-intensive task is first in the dispatch queue, all other tasks with the same service level must wait in the queue until the Load Balancer dispatches the resource-intensive task.
Metric-Based Dispatch Mode
In metric-based dispatch mode, the Load Balancer evaluates nodes in a round-robin fashion until it finds a node that can accept the task. The Load Balancer checks the resource provision thresholds on the first available node. It dispatches the task to this node if dispatching the task causes none of the thresholds to be exceeded. If dispatching the task causes any threshold to be exceeded, or if the node is out of free swap space, the Load Balancer evaluates the next node. It continues to evaluate nodes until it finds a node that can accept the task.
To determine whether a task can run on a particular node, the Load Balancer collects and stores statistics from the last three runs of the task. It compares these statistics with the resource provision thresholds defined for the node. If no statistics exist in the repository, the Load Balancer uses the following default values:
The Load Balancer dispatches tasks for execution in the order the Workflow Manager or scheduler submits them. The Load Balancer does not bypass any tasks in the dispatch queue. Therefore, if a resource intensive task is first in the dispatch queue, all other tasks with the same service level must wait in the queue until the Load Balancer dispatches the resource intensive task.
Adaptive Dispatch Mode
In adaptive dispatch mode, the Load Balancer evaluates the computing resources on all available nodes. It identifies the node with the most available CPU and checks the resource provision thresholds on the node. It dispatches the task if doing so does not cause any threshold to be exceeded. The Load Balancer does not dispatch a task to a node that is out of free swap space.
In adaptive dispatch mode, the Load Balancer can use the CPU profile to rank nodes according to the amount of computing resources on the node.
To identify the best node to run a task, the Load Balancer also collects and stores statistics from the last three runs of the task and compares them with node load metrics. If no statistics exist in the repository, the Load Balancer uses the following default values:
In adaptive dispatch mode, the order in which the Load Balancer dispatches tasks from the dispatch queue depends on the task requirements and dispatch priority. For example, if multiple tasks with the same service level are waiting in the dispatch queue and adequate computing resources are not available to run a resource intensive task, the Load Balancer reserves a node for the resource intensive task and keeps dispatching less intensive tasks to other nodes.
Service Levels
Service levels establish priorities among tasks that are waiting to be dispatched.
When the Load Balancer has more tasks to dispatch than the PowerCenter Integration Service can run at the time, the Load Balancer places those tasks in the dispatch queue. When multiple tasks are waiting in the dispatch queue, the Load Balancer uses service levels to determine the order in which to dispatch tasks from the queue.
Service levels are domain properties. Therefore, you can use the same service levels for all repositories in a domain. You create and edit service levels in the domain properties or using infacmd.
When you create a service level, a workflow developer can assign it to a workflow in the Workflow Manager. All tasks in a workflow have the same service level. The Load Balancer uses service levels to dispatch tasks from the dispatch queue. For example, you create two service levels:
- •Service level “Low” has dispatch priority 10 and maximum dispatch wait time 7,200 seconds.
- •Service level “High” has dispatch priority 2 and maximum dispatch wait time 1,800 seconds.
When multiple tasks are in the dispatch queue, the Load Balancer dispatches tasks with service level High before tasks with service level Low because service level High has a higher dispatch priority. If a task with service level Low waits in the dispatch queue for two hours, the Load Balancer changes its dispatch priority to the maximum priority so that the task does not remain in the dispatch queue indefinitely.
The Administrator tool provides a default service level named Default with a dispatch priority of 5 and maximum dispatch wait time of 1800 seconds. You can update the default service level, but you cannot delete it.
When you remove a service level, the Workflow Manager does not update tasks that use the service level. If a workflow service level does not exist in the domain, the Load Balancer dispatches the tasks with the default service level.
Creating Service Levels
Create service levels in the Administrator tool.
1. In the Administrator tool, select a domain in the Navigator.
2. Click the Properties tab.
3. In the Service Level Management area, click Add.
4. Enter values for the service level properties.
5. Click OK.
6. To remove a service level, click the Remove button for the service level you want to remove.
Configuring Resources
When you configure the PowerCenter Integration Service to run on a grid and to check resource requirements, the Load Balancer dispatches tasks to nodes based on the resources available on each node. You configure the PowerCenter Integration Service to check available resources in the PowerCenter Integration Service properties in Informatica Administrator.
You assign resources required by a task in the task properties in the PowerCenter Workflow Manager.
You define the resources available to each node in the Administrator tool. Define the following types of resources:
- •Connection. Any resource installed with PowerCenter, such as a plug-in or a connection object. When you create a node, all connection resources are available by default. Disable the connection resources that are not available to the node.
- •File/Directory. A user-defined resource that defines files or directories available to the node, such as parameter files or file server directories.
- •Custom. A user-defined resource that identifies any other resources available to the node. For example, you may use a custom resource to identify a specific database client version.
Enable and disable available resources on the Resources tab for the node in the Administrator tool or using infacmd.
Calculating the CPU Profile
In adaptive dispatch mode, the Load Balancer uses the CPU profile to rank the computing throughput of each CPU and bus architecture in a grid. This ensures that nodes with higher processing power get precedence for dispatch. This value is not used in round-robin or metric-based dispatch modes.
The CPU profile is an index of the processing power of a node compared to a baseline system. The baseline system is a Pentium 2.4 GHz computer running Windows 2000. For example, if a SPARC 480 MHz computer is 0.28 times as fast as the baseline computer, the CPU profile for the SPARC computer should be set to 0.28.
By default, the CPU profile is set to 1.0. To calculate the CPU profile for a node, select the node in the Navigator and click Actions > Recalculate CPU Profile Benchmark. To get the most accurate value, calculate the CPU profile when the node is idle. The calculation takes approximately five minutes and uses 100% of one CPU on the machine.
You can also calculate the CPU profile using infacmd. Or, you can edit the node properties and update the value manually.
Defining Resource Provision Thresholds
The Load Balancer dispatches tasks to PowerCenter Integration Service processes running on a node. It can continue to dispatch tasks to a node as long as the resource provision thresholds defined for the node are not exceeded. When the Load Balancer has more Session and Command tasks to dispatch than the PowerCenter Integration Service can run at a time, the Load Balancer places the tasks in the dispatch queue. It dispatches tasks from the queue when a PowerCenter Integration Service process becomes available.
You can define the following resource provision thresholds for each node in a domain:
- •Maximum CPU run queue length. The maximum number of runnable threads waiting for CPU resources on the node. The Load Balancer does not count threads that are waiting on disk or network I/Os. If you set this threshold to 2 on a 4-CPU node that has four threads running and two runnable threads waiting, the Load Balancer does not dispatch new tasks to this node.
This threshold limits context switching overhead. You can set this threshold to a low value to preserve computing resources for other applications. If you want the Load Balancer to ignore this threshold, set it to a high number such as 200. The default value is 10.
The Load Balancer uses this threshold in metric-based and adaptive dispatch modes.
- •Maximum memory %. The maximum percentage of virtual memory allocated on the node relative to the total physical memory size. If you set this threshold to 120% on a node, and virtual memory usage on the node is above 120%, the Load Balancer does not dispatch new tasks to the node.
The default value for this threshold is 150%. Set this threshold to a value greater than 100% to allow the allocation of virtual memory to exceed the physical memory size when dispatching tasks. If you want the Load Balancer to ignore this threshold, set it to a high number such as 1,000.
The Load Balancer uses this threshold in metric-based and adaptive dispatch modes.
- •Maximum processes. The maximum number of running processes allowed for each PowerCenter Integration Service process that runs on the node. This threshold specifies the maximum number of running Session or Command tasks allowed for each PowerCenter Integration Service process that runs on the node. For example, if you set this threshold to 10 when two PowerCenter Integration Services are running on the node, the maximum number of Session tasks allowed for the node is 20 and the maximum number of Command tasks allowed for the node is 20. Therefore, the maximum number of processes that can run simultaneously is 40.
The default value for this threshold is 10. Set this threshold to a high number, such as 200, to cause the Load Balancer to ignore it. To prevent the Load Balancer from dispatching tasks to the node, set this threshold to 0.
The Load Balancer uses this threshold in all dispatch modes.
You define resource provision thresholds in the node properties.