Performance and Scalability for Planning Server

After the initial topology has been agreed upon, Planning Server deployment can be scaled out to meet requirements for business process performance and user workload. The process of scaling out Planning Server is driven from two different perspectives: architecture and application planning.

Business modeling and planning drives how PerformancePoint Planning applications are organized and in turn how they are published. This allows for scaling out based on one or more applications and the chosen application structure and data volume that would be required to support each application. If the application design contains only a single site, then the PerformancePoint Server deployment can only be scaled out to a single computer running Microsoft SQL Server 2005 and a single computer running SQL Server 2005 Analysis Services. However, if the application design contains multiple applications with multiple model sites, then the PerformancePoint Server deployment can be scaled out to multiple SQL Server and Analysis Services servers.

In summary, the business modeling allows for the following:

• Each application can reside on its own computer running SQL Server

• Each model site can be published on its own Analysis Services server

The architectural scale-out process is required to support large user workloads, multiple business process tasks, and complex business process. Front-end Web servers can be deployed in a cluster with Network Load Balancing enabled to ensure that incoming client requests are handled in a timely manner. To provide the required performance for user-requested tasks and actions, multiple servers can be introduced. This will have the largest impact when multiple user actions are submitted simultaneously and the deployed Planning Process Service does not have adequate resources to process all of the submitted requests. Planning Process Service uses the native SQL Server 2005 Service Broker functionality, so optimal SQL Server configuration will also allow Planning Process Service to be used according to the user workload.

While scaling out, you should pay careful attention to the Planning Server configuration settings. Some of the configuration settings, such as connection timeout, will affect the overall performance, even in a well-optimized environment.

For additional information on scaling out, refer to the "Best Practices" documents for SQL Server 2005 and Analysis Services.

Once the deployment is scaled out appropriately, scaling up may be required when small incremental improvements are needed to meet business productivity goals. Scaling up consists of increasing the capacity of each server computer by adding or reconfiguring resources such as hard disks or memory. Changing the operating system or replacing a server with a more capable one will also allow the servers in the deployment to be scaled up. For example, administrators can add processors to turn a single-processor computer into a dual-process computer, or a dual-processor computer into a four-processor computer. Increasing RAM and adding additional disk space to a computer are other methods of scaling up. Using a 32-bit operating system with a maximum of 3 gigabytes (GB) of memory may provide the required level of service. You can also scale up to 64-bit operating systems on servers that have greater than 4 GB of memory.

Planning Server requires no special configuration changes in order to take advantage of scale-up changes.

We recommend that all servers used in a deployment use the same platform. Note that you cannot use clustering in a mixed-platform environment. Windows Server explicitly prevents 32- and 64-bit computers from being clustered together.

For additional information on scaling up, see IIS 6.0 Security Best Practices (IIS 6.0) (https://go.microsoft.com/fwlink/?LinkId=102487&clcid=0x409).

Productivity for business users is driven by their ability to use Planning Server to complete business tasks and operations. To keep Planning Server running and to protect against unforeseen hardware issues, various strategies can be deployed:

• You can deploy front-end Web servers in a cluster with Network Load Balancing, which enables the PerformancePoint Server system to allow for high availability.

• You can deploy computers running SQL Server in a clustered environment, which enables the applications to allow for high availability.

• You can deploy Analysis Services and Analysis Servers in a clustered environment, which enables the model sites and models to allow for high availability.

• You can implement RAID 5.0 or 6.0. Mirroring SQL Server and Analysis Services server will also allow for data redundancy and support high availability.

Having multiple asynchronous servers in the deployment is highly recommended. However, there is no clustering support for these servers as Windows Server 2003 does not allow clustering of computers running the same Windows service.

All of the client interactions with the server are performed over the network except when PerformancePoint Server is installed on a stand-alone deployment. In a distributed deployment, the network capacity dictates the speed at which the business data and metadata are moved around. The flow of data from Planning Web Service computers to SQL Server databases primarily consists of movement of metadata, reference data, and fact data. The data flow from SQL Server databases to Analysis Services databases consists of transactional and planning data as well as the related security settings. The latter data flow by far outweighs the data flow from front-end Web servers to SQL Server databases, because publishing data to Analysis Services is a more frequent operation.

Planning Server offers both online and offline modes of operation to support planning processes. While in offline mode, PerformancePoint Add-in for Excel retrieves and stores the dataset on client computers based on the user's security configurations. Therefore network load is also determined by the security configurations defined for business users. Security configurations must be defined in Planning Business Modeler with the smallest scope to ensure that the minimum amount of data is transmitted across the network. If security is not configured appropriately, clients may experience delays in data retrieval and refresh, which would affect the offline functionality in PerformancePoint Add-in for Excel.

We highly recommend that administrators perform appropriate online and offline performance benchmarking in design and test environments to ensure that the deployed environment will provide the required level of service to business users.

User workload in Planning Server results from the application modeling and design activities in Planning Business Modeler as well as the business process and data submissions from PerformancePoint Add-in for Excel. Much of the user workload generated by the clients is submitted to the server over the network as Web service requests encapsulating the data and the business operations. The dataset size and the related business operations also contribute to the overall user workload.

Scaling out the Planning Server deployment will likely allow for larger workloads to be handled. We highly recommend that Planning Server administrators perform basic benchmarking in design and test environments to ensure that the deployed environment will provide the required level of service to business users.

Planning Server is deployed in two stages: software installation and software configuration.

First, Planning Server is installed on the computer. Everything necessary to run and configure Planning Server is available for configuration.

The two configuration options are:

• Stand-alone: This option configures all Planning Server components on one computer, including Planning Server databases. To run complete configuration, SQL Server must be installed on the target computer.

• Distributed: This option configures one, two, or all Planning Server components. This is the option that allows Planning Server and Planning Server databases to be on separate computers. In a distributed, multi-server topology, you may have to perform the custom configuration process on multiple computers.

The configuration stage of the process consists of Planning Server Configuration Manager configuring each Planning Server computer. Note that the client installations, Planning Business Modeler and PerformancePoint Add-in for Excel, perform installation and configurations in a single step.

Planning Server Configuration Manager automatically runs after the initial server installation. Using Planning Server Configuration Manager, choose the servers that you want to configure. Examples include configuring the Web sites and the computers that run SQL Server.

Planning Server Configuration Manager can be run multiple times. For example, you may configure Planning Web Service and later return to configure the Remote Administration Service.

Planning Server Stand-alone installation

Planning Server distributed installation

Think about the following interoperability considerations for deploying Planning Server in a distributed installation. They are requirements for ensuring a successful deployment.

The installation of Planning Server computers must be mirrors of each other. This means that any code or component installed on a Planning Server computer must be installed on all Planning Server computers in a Web farm. For example, if Planning Web Service is installed on one computer in a Web farm, it must be installed on all computers.

All Planning Server computers are stateless for better farm support. From the highest level, this means that any Planning Server computer in a Web farm can be replaced by a similar server without loss of server configuration data and committed user data. This requirement should not be confused with the application being stateless. The only data that may have been lost would be in-process data which had not been committed prior to failure.

There are cases where multiple SQL Server relational databases and SQL Analysis Services databases are not only allowed, but encouraged. These include:

• When you have large data stores

• When one or more services places large demands on an instance of SQL Server or Analysis Services

This topic is included in the following downloadable book for easier reading and printing:

• Planning Server Planning and Architecture Guide for PerformancePoint Server 2007

See the full list of available books at Downloadable content for PerformancePoint Planning Server.