Database Reference
In-Depth Information
falls into, and the second level concerns the
location of the element within the block.
The chunking process is equivalent to the
block partitioning method used for matri-
ces. For extremely large multidimensional
arrays, a B+ tree storage scheme can be
used for the first level chunk organization.
The accessing of array elements in multi
levels is also used in EKMR (Chun et, al;,
2003) scheme when the number of dimen-
sion is more than four. Kaser & Daniel,
(2003) and Owen (2002) determine some
ways to store data cubes using different
coding for dense regions than for sparse
ones. A single dense sub-cube is found and
the remainder is considered sparse.
Compressed implementation of flexible
•
Applying indexing techniques on com-
pressed data warehouses. The storage re-
quirement for the indices will also be re-
duced because the size of the database has
been reduced by compression. Bitmap in-
dices are designed for different query types
including range, aggregation and join que-
ries. The basic idea is to indicate whether
an attribute in a tuple is equal to a specific
value or not (O'Neil & Quass, 1997) and
Chan & Ioannidis (1998). Queries whose
predicates involve attributes indexed by
bitmap indexes can be answered fast by
performing bitwise AND, or OR, or NOT
operations on bitmaps, which is a big ad-
vantage of bitmap indexes. The size of a
bitmap index strongly depends on the car-
dinality of an indexed attribute. For attri-
butes of high cardinality, indexes would
be very large and hence the performance
of data access with the support of such in-
dexes might deteriorate (Wu & Buchmann,
1998). The dynamic bitmap index is con-
structed dynamically (Sarawagi, 1997)
using vertical partition of the table, where
each column stores a compressed repre-
sentation of the values in the correspond-
ing indexed column. In bitmap join index
(Vanichayobon & Gruenwald, 1999) an
index is created for a join attribute of two
tables so that the actual join needs not be
performed. This is effective for low cardi-
nality data. Considering different criteria,
a data cube is partitioned and compressed
by Buccafurri et, al; 2003 by substituting
each block with a few summary informa-
tion. Three 64-bits indexing techniques for
compressed data warehouses are present-
ed, namely 2/3LT, 2/4LT and 2/pLT index-
es (LT stands for Level Tree). The 2/3LT
index is balanced and suitable for distribu-
tions with no strong asymmetry, the 2/4LT
is for biased for a distribution, and 2/pLT
(p for peak) captures distribution having
•
and extendible MOLAP, so that the dy-
namic treatment of the array becomes fea-
sible. In both scientific and MOLAP data
storage, the data grows incrementally over
time and as such the array storage map-
ping must be extendible (Rotem & Zhao,
1996 and Tsuji et, al; 2008). Hence, incre-
mental maintenance of the data cube is an
important future MOLAP implementation
and research issue (Kim & Lee, 2006 and
Jin et, al; 2008). The incremental mainte-
nance of the data cube means the propaga-
tion of its changes. Generally the amount
of changes during the specified time is
much smaller than the size of the source
relation. Therefore, it is necessary to com-
pute only the changes of the source relation
and reflect this into the original data cube.
Kim & Lee (2006) introduces incremental
data cube maintenance for ROLAP and Jin
et, al; 2008 employed the same approach
for MOLAP data cube computation. A
MOLAP array is extendible if the array
bounds are allowed to grow by admitting
new array elements that are appended to
the storage space but without reorganizing
previously allocated elements.
Search WWH ::
Custom Search