New Data Warehouse Technologies - Data Warehouse Systems: Design and Implementation - page 506

a

RowId EmployeeKey CustomerKey ProductKey ···

1 e1 c1 p1 ···

2 e1 c1 p4 ···

3 e1 c2 p4 ···

4 e1 c2 p4 ···

5 e1 c2 p4 ···

6 e2 c2 p5 ···

7 e2 c2 p5 ···

8 e2 c2 p1 ···

9 e3 c3 p2 ···

10 e3 c3 p2 ···

··· ··· ··· ··· ···

b c d

f v l f v l f v l

1 e1 5 1 c1 2 1 p1 1

6 e2 3 3 c2 6 2 p4 4

9 e3 2 9 c3 2 6 p5 2

... ... ... ... ... ... 8 p1 1

... ... ... ... ... ... 9 p2 2

... ... ... ... ... ... ... ... ...

Fig. 13.3 Storing columns of a fact table one table per column. ( a ) Fact table Sales .
( b )Column EmployeeKey .( c )Column CustomerKey .( d )Column ProductKey

13.4 In-Memory Database Systems

An IMDBS is a DBMS that stores data in main memory, opposite to
traditional database systems, which store data on persistent media such as
hard disks. Because working with data in memory is much faster than writing
to and reading from a file system, IMDBSs can run applications orders of
magnitude faster. IMDBSs come in many flavors: they can be DBMSs that
only use main memory to load and execute real-time analytics, they can
be used as a cache for disk-based DBMSs, or they can be commercialized
as software-hardware licensed packages, called appliances, particularly for
business intelligence applications. In most cases, they are combined with
column-store technology.
The typical way in which traditional DBMSs operate is based on reading
data from disk to buffer pages located in main memory. When a query is
submitted, data are first fetched in these buffers, and, if not found, new data
are loaded from disk into main memory. If there is an update, the modified
page is marked and written back to disk. The process where disk-based
databases keep frequently accessed records in memory for faster access is