SAS SOFTWARE COMPONENTS USED FOR THE IMPLEMENTATION

The following sections clarify the SAS software components used and the fine points of some technical information around such a grid installation.

SAS GRID MANAGER

SAS Grid Manager was introduced in SAS 9.1.3. It builds upon the parallel capabilities of SAS/CONNECT and adds many other capabilities required by enterprise grid deployments. SAS Grid Manager provides load balancing, policy enforcement, efficient resource allocation, and prioritization for SAS products and solutions running in a shared grid environment. In addition, it de-couples the SAS applications and the infrastructure used to perform the applications.

This allows hardware resources to transparently grow or contract as needed and provides tolerance of hardware failures inside the grid infrastructure. SAS Grid Manager integrates the resource management and scheduling capabilities of the Platform Suite for SAS with the SAS 4GL syntax and then with several SAS products and solutions. This integration allows you to create a managed, flexible, and common environment to efficiently process the following to meet your business needs:

1) Multiple users

2) Parallel workloads

3) Enterprise scheduling

Multiple Users

A lot of clients have a number of ad hoc SAS users that develop models, do queries, and other sorts of ad hoc development, discovery and analysis. SAS Grid Manager provides management of this ad hoc environment from the viewpoint of controlling which jobs get priority over others, which jobs get a particular share of the computing resources, and the prevention of mass submission of work requests that result in everyday server crashes. SAS Grid Manager will map the SAS work requests to the available resources, and if necessary, run just a subset of the work and queue the remaining work requests for execution as soon as resources turn out to be available. High priority jobs can even preempt lower priority work so that the most critical business processes execute first. SAS Grid Manager provides the structure needed to create an organized and managed SAS analytic environment to ensure that the appropriate property are allocated to the appropriate users and that the workload meets the objectives of the organization.

The total SAS user community is likely using many different interfaces for running SAS applications. Some portion of the users might be running SAS in interactive Display Manager System (DMS) mode. In this case, jobs can be interactively submitted to a grid environment by important a new key sequence to submit the appropriate statements to send the job to the grid rather than executing on the local workstation. Similarly, for those users who prefer group job submissions, a wrapper script file can be used to submit batch jobs to the grid with no change to the application and no change in the way your users interact with SAS. Additionally, SAS Enterprise Guide users can submit their SAS applications and SAS Enterprise Guide tasks to a SAS grid for execution. SAS Grid Manager provide the infrastructure to balance all SAS applications and workload across a shared grid infrastructure.

Parallel Workloads

A subset of SAS applications consist of sub-tasks that are independent units of work and can be distributed across a grid and executed in parallel. The benefit of distributed parallel execution is potentially substantial acceleration of the entire application.

A common workflow in applications created by SAS Data Integration Studio is to repeatedly execute the same analysis against different subsets of data. For this workflow, the Loop and Loop-End transformation nodes can be used in SAS Data Integration Studio to automatically generate a SAS application that will spawn every iteration of the loop to a SAS grid via SAS Grid Manager.

SAS Risk Dimensions has a similar iterative workflow of executing the same analysis over different subsets of data. In the case of SAS Risk Dimensions, the data is subset based on market states or by instruments, and every iteration of the analysis can be submitted to the grid using SAS Grid Manager to provide load balancing and well-organized resource allocation.

In contrast, the workflow for SAS Enterprise Miner during the model training phase is to execute a series of different models against a common set of data. Because the models are independent of each other, the SAS Enterprise Miner component that generates the SAS program to execute the user-created flow will automatically insert the necessary syntax to spawn each model execution to the grid for parallel and ultimately accelerated execution. Finally, a programmer can modify an obtainable application or develop a new application that consists of replicate runs of the same fundamental task or multiple distinct independent units of work. In this case, the programmer can use the grid syntax available in the SAS 4GL programming language to create a grid-enabled application. In all of these scenarios, SAS Grid Manager will hand out and load balance the parallel work requests to a shared pool of SAS grid resources.

Enterprise Scheduling

Scheduling production jobs is an important and necessary function in every enterprise. SAS provides the Schedule Manager plug-in as part of the SAS Management Console to enable you to create SAS workflows and to list them based on time and file events, or both. SAS Grid Manager will then distribute the jobs within one or more workflows to a SAS grid environment for load balancing and sharing of resources. The SAS jobs can be created by a diversity of SAS applications and solutions as well as written by SAS programmers. Various levels of scheduling capabilities have been incorporated directly into many SAS products and solutions including SAS Data Integration Studio, SAS Marketing Automation, SAS Marketing Optimization, and SAS Web Report Studio. The jobs created by these products, as well as any other SAS products, including user-written SAS programs, can be used to create simple or very complex workflows and planned to a SAS grid environment.

The use of SAS Grid Manager Software enables workload stemming from data integration, reporting, and solution clients to be dynamically and transparently routed to the least used node at this time. That can be any host participate in the grid, regardless of whether the host is located in a local or remote site as long as it can be reached over the network. This ensures that a client’s request will be executed in the shortest achievable time. In fact, it guarantees that the request will be run, even when one or more grid nodes are not obtainable due to ongoing protection or even a hardware failure.