PI Integrator for Business Analytics scale architecture

The following diagram shows the PI Integrator for SAP HANA architecture.

PI Integrator for Business Analytics with one PI Integrator Worker Node

The PI Integrator Framework, PI Integrator Worker Node, and PI Integrator Sync processes all reside on the same computer.

Each PI Integrator for Business Analytics instance has a minimum of one worker node process that manages the publishing of views including scheduled jobs, streaming jobs, and run once jobs. You can add additional worker nodes to improve the publishing performance. This can be done during installation of PI Integrator for Business Analytics. After installation, you can add more nodes using the Add Remove Programs option in the Microsoft Windows Control Panel.

Note that each additional worker node has additional RAM and CPU requirements. For more information, see System requirements. You will not be able to add nodes from the PI Integrator for Business Analytics web application.

The PI Integrator Worker Node publishes jobs to the destination targets. The Cluster Manager automatically distributes jobs, in turn, among the available nodes. For example, assume there are two worker nodes and five jobs. The first job is assigned to the first worker node, the second job is assigned to the second worker node, the third job is assigned to the first worker node, and so on. When a worker node is added to the cluster, all PI Integrator for Business Analytics services are restarted and the Cluster Manager balances the jobs across all available nodes, including the newly added node.

Note: Jobs are not assigned to the nodes in alphabetical order. If a node goes down, the Cluster Manager reassigns any jobs that were assigned to it to the remaining worker nodes. The offline worker node is automatically restarted by the PI Integrator Framework service. Once restarted, existing jobs are not rebalanced to include the restarted worker node. However, new jobs are balanced across all working nodes. If this results in an unbalanced distribution of jobs, you can manually rebalance the jobs.

Optimizing the worker node processes

Follow these guidelines to optimize the performance of your worker nodes:

• Install a minimum of two PI Integrator Worker Nodes, ensuring that you have the required RAM and CPU to support additional worker nodes. Then, if one of the nodes fails, the other will be able to pick up the jobs from the failed node and view data will still get published.

• Review your view statistics and identify which views are taking so long to publish that the next scan is missed. Break up problematic views into multiple smaller views. If the problem persists, then consider adding another node.

• Check that your views are efficiently structured. Use PI AF templates whenever possible, but do not create shapes that return bloated results. The number of shape matches has a direct impact on performance. Therefore, if possible, design your search shape on the Select Data page of creating a view to specify only those elements you need and filter out unnecessary elements. (This is before any row filters are applied.)

• Schedule your views so that they are not all publishing at the same time. Offset the jobs from one another.

If, after optimizing your views, you still need to improve the performance, then add a worker node to the cluster. See System requirements for the memory and hardware requirements. You can have up to five PI Integrator Worker nodes. All worker nodes must reside on the same computer on which the PI Integrator for Business Analytics is installed.

PI Integrator for Business Analytics with multiple PI Integrator Worker nodes