I have reviewed much of the overview presentation (http://www.sdn.sap.com/irj/scn/go/portal/prtroot/docs/library/uuid/3604c604-0901-0010-f0aa-b37378495537?quicklink=index&overridelayout=true) but I am confused about slide 43 that describes "BW Accelerator Architecture: Vertical Decomposition."
I am trying to learn more about this appliance but I can't wrap my brain around the storing tables by row vs. by column idea. I understand it on its face: storing by row requires more steps to search the data. What I don't understand is why precisely this ends up being faster to store by column. Additionally - how exactly do you store the data by column - if the attributes are separated in the database (rather than being stored as tuples) then how does the database reconnect a particular column's attribute values with the row values?
I appreciate the help, thank you in advance.