Physical Memory (RAM) Configuration
icCube being a in-memory OLAP server, you must pay attention to the configuration of the amount of physical RAM allocated for the Java Virtual Machine (JVM).
Facts (aka. measures) Files
Since version 2.5.6, icCube adds more flexibility and allows for storing facts into files to minimize the actual amount of physical memory required (Enterprise II). This option is available on both Windows and Linux. We highly recommend Linux where the I/O are faster, steadier and do not suffer of files being locked. If used on Windows with an Oracle JVM, we recommend using a JRE 1.7 (instead of 1.6).
New configuration properties are available the icCube.xml file: icCube.factFileSize / icCube.factFileBufferSize. See comments in this file for detailed and up-to-date explanations about their purposes and actual values.
This option is defined at schema level as an advanced property. All cubes from the schema will store their facts into files:
Use Case : Lots of Cubes
You are having several cubes, each one fitting in RAM but the amount of physical RAM is not enough to load all the cubes in RAM.
You can then setup several (or all) cubes to have their facts stored in files. Performance (that is MDX query processing time) is then depending on how often each cube is being accessed by the MDX queries. For example, if the same cubes are re-used very often then they will be cached in RAM by the OS and you should not see any performance issue compared to a full RAM configuration.
Use Case : Large Cube
You are having a large cube that does not fit in RAM.
Storing the facts into files will solve this issue. Performance (i.e., MDX query processing time) is now depending on how many facts you are going to access as this might require accessing files. But if the queries are accessing sub-cubes (e.g., drilldown, filtering, etc...), then it is more likely that facts are going to be cached by the OS within the amount of physical RAM your system has (the more the better) and you should not see any performance issue compared to a full RAM configuration.