|Oracle9i OLAP Services Concepts and Administration Guide
Release 1 (9.0.1)
Part Number A88755-01
Tuning, 3 of 5
OLAP Services will perform best if it is running on a new computer dedicated to its use. If your budget allows for this expenditure, then you should invest in new equipment. A computer that is only a year or two old is significantly slower than one that you purchase today. While this may appear to be an expensive solution, it can more than pay for itself in labor savings. Your time costs money.
The computer should have at least two CPUs. Acquire the fastest CPUs available, even if that means having fewer of them.
Note: If you must use a computer with only one processor, then you need to change the FreePageHighCount, FreePageLowCount, and related configuration settings, since the default configuration assumes two CPUs. Search for "FreePageLowCount" in Instance Manager Help for guidelines.
The amount of memory used by OLAP Services varies widely depending upon the type of application being used.
A strict relational OLAP application (typically called ROLAP) needs 16 MB per instance with the PageBufferCount configuration parameter set to 4096 buffers. ROLAP applications use OLAP Services as a temporary cache for data that is selected and calculated in the database.
A multidimensional OLAP application (typically called MOLAP) can require significantly more resources, particularly if large amounts of data are being aggregated in the analytic workspace. On a 32-bit computer, OLAP Services can use up to 4 GB of memory per instance. On a 64-bit computer, OLAP Services can use as much memory as you provide. In determining how much to add, you can consider, in addition to your budget, the size of the persistent workspace or its largest variable.
If OLAP Services is installed on the same computer as the relational database, then its requirements must be added to those of the database. Moreover, you must set the PageBufferCount configuration setting to no more than half of physical memory to allow a sufficient amount for use by the database. Use the following equation for this calculation:
On 32-bit computers, limit the result to a maximum of 524,288 buffers.
While you are evaluating your hardware requirements, be sure to include your network. Users of business intelligence applications can significantly increase the traffic on an existing network, since they typically access large volumes of data. If there is a bottleneck at any point between the server and the end user, then performance will suffer.