Hardware Recommendations

As you plan your IBM® SPSS® Modeler Server installation, you should consider the hardware that you will use. Although IBM SPSS Modeler Server is designed to be speedy, you can maximize its efficiency by using hardware that is sized appropriately for your data mining tasks. Upgrading hardware is often the simplest and most economical way to improve performance across the board.

Dedicated server. Install IBM SPSS Modeler Server on a dedicated server machine where it will not compete for resources with other applications, including any databases to which IBM SPSS Modeler Server may be connecting. Model-building operations in particular are resource-intensive and perform much better when not in competition with other applications.

Note: Although installing IBM SPSS Modeler Server on the same computer as the database can reduce data-transfer time between the database and the server by avoiding network overhead, in most cases the best configuration is to have the server and database on separate machines to avoid competition for resources. Provide a fast connection between the two to minimize the cost of data transfer.

Processors. The number of processors on the machine should be no less than the number of concurrent tasks (simultaneously executing streams) you expect to run on a regular basis. In general, the more processors, the better.

  • A single instance of IBM SPSS Modeler Server will accept connections from multiple clients (users), and each client connection can initiate multiple stream executions. One server can therefore have several execution tasks in progress at any one time.
  • As a rule of thumb, allow one processor for one or two users, two processors for up to four users, and four processors for up to eight users. Add one additional processor for every two to four users beyond that, depending on the mix of work.
  • To the extent that some processing may be pushed back to the database through SQL optimization, it may be possible to share a CPU between two or more users with minimal loss in performance.
  • Multithreading capabilities make it possible for a single task to take advantage of multiple processors, so adding processors can improve performance even in cases where only one task is running at a time. Generally, multithreading is used for C5.0 model building and certain data preparation operations (sort, aggregate, and merge). Multithreading is also supported for all nodes that run in IBM SPSS Analytic Server (for example: GLE, Linear-AS, Random Forest, LSVM, Tree-AS, Time Series, TCM, Association Rules, and STP).

64-bit platforms. If you plan to process or build models on very large volumes of data, use a 64-bit machine as your IBM SPSS Modeler Server platform, and maximize the amount of RAM for the machine. For larger data sets, the server can quickly exhaust the per-process memory limits imposed by 32-bit platforms, forcing data to be spilled to disk and significantly increasing the running time. 64-bit server implementations can take advantage of additional RAM; a minimum of 8 gigabytes (GB) is recommended.

Future needs. Whenever feasible, make sure that server hardware is expandable in terms of memory and CPUs, both to accommodate increases in usage (for example, increased numbers of simultaneous users or increases in the existing users processing requirements) and increased multithreading capabilities of IBM SPSS Modeler Server in the future.