For optimal session performance, configure the cache size so that the Integration Service processes the transformation in memory without paging to disk. Session performance decreases when the Integration Service pages to disk.
When you use the cache calculator to calculate the cache size, the cache calculator estimates the cache size required for optimal session performance based on your input. You can tune the estimate by using the cache size specified in the session log. After you run the session, review the transformation statistics in the session log to get the cache size.
For example, you run an Aggregator transformation called AGGTRANS. The session log contains the following text:
MAPPING> TT_11031 Transformation [AGGTRANS]:
MAPPING> TT_11114 [AGGTRANS]: Input Group Index = , Input Row Count 
MAPPING> TT_11034 [SQ_V_PETL]: Input - 110264
MAPPING> TT_11115 [AGGTRANS]: Output Group Index = 
MAPPING> TT_11037 [FILTRANS]: Output - 1098,Dropped - 0
MAPPING> CMN_1791 The index cache size that would hold  aggregate groups of input rows for [AGGTRANS], in memory, is  bytes
MAPPING> CMN_1790 The data cache size that would hold  aggregate groups of input rows for [AGGTRANS], in memory, is  bytes
The log shows that the index cache requires 286,720 bytes and the data cache requires 1,774,368 bytes to process the transformation in memory without paging to disk.
The cache size may vary depending on changes to the session or source data. Review the session logs after subsequent session runs to monitor changes to the cache size.
You must set the tracing level to Verbose Initialization in the session properties to enable the Integration Service to write the transformation statistics to the session log.
The session log does not contain transformation statistics for a Sorter, a Joiner transformation with sorted input, an Aggregator transformation with sorted input, or an XML target.