New Data Warehouse Technologies - Data Warehouse Systems: Design and Implementation - page 508

Main memory

Active data

Main store Buffer store Additional structure

Indexes
Object guides
Prejoins
...

Merge

Columns (single and
combined)

Data aging
algorithms

Recovery Logging

Historic data
(passive)

Database snapshots Transaction log

Traditional database

Fig. 13.4 A typical IMDBS architecture

for fast retrieval of low cardinality data, like primary key columns, or object
data guides (in the case of SAP), which allow complex data objects stored
as a hierarchy of elements to be reconstructed. Finally, although data in the
database are stored in main memory, to save memory space, IMDBSs also
store data persistently. This is done as follows. The most recent data are
kept in main memory, since these are the data most likely to be accessed
and/or updated. These data are called active . Opposite to this, passive
data are data not currently used by a business process, used mostly for
analytical purposes. Passive data are stored on nonvolatile memory, even
using traditional DBMSs. This supports so-called time-travel queries, which
allow the status of the database as of a certain point in time to be known.
Data partition between active and passive data is performed by data aging
algorithms. Nonvolatile memory is also used to guarantee consistency and
recovery under failure: data updates are written in a log file, and database
snapshots are kept in nonvolatile memory, to be read in case of failure. This
combination of main and nonvolatile memory is supposed to allow IMDBSs to
support OLTP transactions and OLAP analysis at the same time. However,
this capability is currently being questioned by researchers and practitioners.