Hi Jagaa,
To answer part of your questions; actually HANA persists data physically on disk and only what you are querying against will be loaded into memory. So lets say you have 100 tables replicated into HANA and stored on hard disk. Somebody runs a query against 1 of the tables, lets say it's a customer table. If the customer table has 20 columns (ie: fields) in it yet the query is only querying 2 fields such as customer number and customer name then only those two columns are loaded into memory. (plus any key columns).
As far as when memory gets full; when memory reaches a certain threshold HANA will automatically start to unload some of the oldest/least used columns based on an algorithm and different factors. In fact you can configure certain tables to have different UNLOAD_PRIORITY so that they are unloaded first.
This is all greatly simplified and I could go into a lot more detail on the delta process etc but I hope this answers some of your questions. I'm not a compression expert so I will let somebody else answer that.
-Patrick