There is a concept of hot/warm/cold data in a datawarehouse.
- Hot refers to the data frequently accessed in reports like recently loaded request in billing cube.
- Warm refers to the data that is used not so frequently but is accessed time and again in the reports like customer master data.
- Cold refers to the data that is sparingly used in reporting like redundant data residing in PSAs, older than 3 years of data in targets etc.
For SAP BW on HANA database, only the hot and warm data should be kept in memory and cold data should be off loaded from memory. This is required because HANA memory is limited and should be efficiently utilized.
But again, sometimes the HANA memory becomes full of the hot/warm data too and some of it should be removed from memory. In such cases, the HANA system moves some of the data from the memory to the HANA file system.
This is called as early unload concept in SAP BW powered by HANA database.
Whenever a new request is loaded like in PSA or a dso, it is stored in a new partition by the BW system. The HANA system selectively identifies the old partitions and off-loads them from memory.
To do these offloads in the HANA system, you can set a flag for the objects to direct the HANA system to offload them first from the memory. This can be done in the tcode RSHDBMON.
Go to this tcode and search for the objects in say an InfoArea. Select all the tables associated with your object from the list shown and click on the LOAD/ UNLOAD icon. Then select ‘Activate Early Unload‘ option.
Please note: The write optimized dsos based on 3.x data flow does not use partition id and hence full data is loaded to memory. Thus, there in no advantage in flagging them as ‘early unload’.
Also the write optimized dsos with the setting ‘Duplicate data records not allowed’ are not partitioned by request ids and hence full data is loaded to memory. Here also, the ‘early unload’ flag concept will not work.