use sparse lookup when reference database is huge i.e ratio 
of input data and reference data is 1:100 or more

If reference data is huge then we will go for join stage. 
Then when we will use join and sparce is useful?

File set can be stored on multiple Unix files as flat files.
 file set contains discriptor file and individual raw data 
files . No. of raw data files depends on configuration file.

There will be some limit like 2G limit on some file 
systems . we can distribute data over nodes to prevent 
overrun.
In such cases file set will be useful than sequential file.

First we Know about Seq file and Dataset.

Seg file is used Extract the from flatfiles and load into 
flatfiles and limit is 2GB

Dataset is a intermidiate stage and it has paralism when 
load data into dataset and it inprove the performanece.

combinality is a mode in Advance tab of a stage.

The mode can have values: Auto, Combinable, Don't Combine. 
Auto: let the DataStage decide whether to combine the 
stages 
Combinable: Combine the operator with other stage if 
possible 
Don't Combine: do not combine them at all 

I think Don't Combine may be non combinability in your 
terms.