Data Load MAXLs in ASO

Question

Hi All,
Greetings of the day !!!!
Want to understand the difference between "Add values create slice" and "override values create slice" used in data loading MAXLs
Suppose we initialized buffer and loaded data in buffer then we can use following two MAXLs
1)
import database AsoSamp.Sample data
from load_buffer with buffer_id 1
add values create slice;
2)
import database AsoSamp.Sample data
from load_buffer with buffer_id 1
override values create slice;
Q1
What i am thinking logically is if i am again loading the data in the same intersections from which slice is created ADD VALUE will add it and override value will overwrite it .... e.g if 100 was present earlier and we are again loading 200 then ADD will make 300 and overwrite will result 200. Let me know if my understanding is correct
Q2
Why do we use "create slice" ? What is the use? Is it for better performance for data loading? Is it compulsary to merge the slices after dataloading??
Cant we just use add value or override values if we dont want to create slice...
Q3
I saw two MAXLs for merging also ... one was Merge ALL DATA and other was MERGE incremental data ... Whats the diff ? In which case we use what?
Pls help me in resolving my doubts... Thanks a lot !!!!

Answer 1

Q1 - Your understanding is correct. The buffer commit specification determines how what is in the buffer is applied to what is already in the cube. Note that there are also buffer initialization specifications for 'sum' and 'use last' that apply only to data loaded to the buffer.
Q2 - Load performance. Loading data to an ASO cube without 'create slice' takes time (per the DBAG) proportional to the amount of data already in the cube. So loading one value to a 100GB cube may take a very long time. Loading data to an ASO cube with 'create slice' takes time proportional to the amount of data being loaded - much faster in my example. There is no requirement to immediately merge slices, but it will have to be done to design / process aggregations or restructure the cube (in the case of restructure, it happens automatically IIRC). The extra slices are like extra cubes, so when you query Essbase now has to look at both the main cube and the slice. There is a statistic that tells you how much time Essbase spends querying slices vs querying the main cube, but no real guidance on what a 'good' or 'bad' number is! See http://docs.oracle.com/cd/E17236_01/epm.1112/esb_tech_ref/aggstor_runtime_stats.html.
The other reason you might want to create a slice is that it's possible to overwrite (or even remove, by committing an empty buffer with the 'override incremental data' clause in the buffer commit specification) only the slice data without having to do physical or logical clears. So if you are continually updating current period data, for example, it might make sense to load that data to an incremental slice.
Q3 - You can merge the incremental slices into the rest of the cube, or you can merge multiple incremental slices into one single incremental slice, but not into the rest of the cube. Honestly, I've only ever wanted to use the first option. I'm not really sure when or why you would want to do the second, although I'm sure it's in there for a reason.