url - Universität zu Lübeck

4.3. HYBRID APPROACHES 69
paths when processing a path expression. Two functions may be evaluated for
an extent e: The first Boolean function governs(e, v) responds if the extent or one
of its descendant contains an element with the requested value v. The second
Boolean function governs(e, v) returns only true, if e itself contains v. In the sample
data only the node with the id 5 contains an element with the value Singapur,
but all nodes with ids 0 to 5 govern the value.
The authors propose two methods to capture the content in the extents: The first
approach assigns a unique id to every value in the XML data and puts it into
the extents that contain/govern this value. Because an own id for each values
increases the size of the DataGuide dramatically the second proposal uses binary
signatures of a restricted length. A non-bijective function assigns values to signatures.
If an extent governs or contains more than one value the corresponding
signatures are unified bitwise to a single signature that represents all values. This
process is not lossless, leads to false positives and therefore requires postprocessing
when evaluating a path expression.
A major issue of the CADG is its limited capability to deal with updates. When
adding/deleting a node or when changing the value of a node the corresponding
signatures/ids must be identified and recalculated respectively deleted. In general,
an update implies that all extents of the Content-Aware DataGuide must be
touched. This is a linear complexity in the size of the database.
4.3.3 ViST
With the Virtual Suffix Tree (ViST) [114] Wang et al. introduce an approach that
encodes and represents XML data and path expressions as structure-encoded
sequences. XML data is represented by the preorder sequence of its tree structure
produced by a depth-first traversal of the XML data. The value of elements
and attributes and the labels of all elements are combined to one large sequence.
Therefore, ViST is comparable to a numbering schema. Since isomorphic trees
may produce different preorder sequences an order among sibling nodes is enforced
using the lexicographic order of the labels. Multiple siblings with the same
label (e.g. the payment element in the XMark sample data) are ordered randomly.
In order to motivate the ViST approach we use the following DOM-represented
XML fragment in figure 4.10.
ViST transforms an XML data into a sequence of (symbol, prefix) pairs - the socalled
structure-encoded sequence D. The XML fragment of figure 4.10 leads to
the following structure-encoded sequence D:
1 D=
2 ( , ) ,
3 (,) ,
4 (,) ,
5 (,) ,
6 (,) ,