1 Indexing and Querying XML Data for Regular Path Expressions A Paper by Quanzhong Li and Bongki Moon Presented by Amnon Shochot

2 Our Objective Developing a system that will enable us to perform XML data queries efficiently.

3 XML Queries Languages Used for retrieving data from XML files. Use a regular path expression syntax. e.g. XPath, XQuery.

4 Queries Today - Inefficient Usually XML tree traversals – Inefficient. –Top-Down Approach –Bottom-Up Approach –An example: the query: /chapter/_*/figure (finding all figures in all chapters.)

5 Our Objective - Refined Developing a system that will enable us to perform XML data queries efficiently Developing such a system consists of: –Developing a way to efficiently store XML data. –Developing efficient algorithms for processing regular path expressions (e.g. XQuery expressions).

6 Storing XML Documents Question: What would we need from a data structure to be able to perform an efficient query? Answer: A mechanism for: –Efficiently finding all elements/attributes with a given name. –Efficiently finding all values with a given name. –Efficiently resolving ancestor-descendant relationship.

7 Storing XML Documents - XISS XISS - XML Indexing and Storage System. Provides us with ways to: –efficiently find all elements or attributes with the same name string grouped by document which they belong to. –quickly determine the ancestor-descendant relationship between elements and/or attributes in the hierarchy of XML data hierarchy.

8 Determining Ancestor-Descendent Relationship According to Dietz’s: for two given nodes x and y of a tree T, x is an ancestor of y iff x occurs before y in the preorder traversal and after y in the postorder traversal. Example:

9 Determining Ancestor-Descendent Relationship – cont. Advantage: the ancestor-descendent relationship can be determined in constant time. Disadvantage: a lack of flexibility. –e.g. inserting a new node requires recomputation of many tree nodes.

10 A new numbering scheme: –Each node is associated with a pair: For a tree node y and its parent x: [order(y), order(y) + size(y)]  (order(x), order(x) + size(x)] For two sibling nodes x and y, if x is the predecessor of y in preorder traversal holds: order(x) + size(x) < order(y). Determining Ancestor-Descendent Relationship – cont. exclusive

11 Determining Ancestor-Descendent Relationship – cont. Fact: for two given nodes x and y of a tree T, x is an ancestor of y iff: order(x) < order(y)  order(x) + size(x)

12 Determining Ancestor-Descendent Relationship – cont. Properties: –the ancestor-descendent relationship can be determined in constant time. –flexibility – node insertion usually doesn’t require recomputation of tree nodes. –an element can be uniquely identified in a document by its order value.

13 XISS System Overview

14 XISS System Overview How the system works: –XML documents are loaded into the XISS system. –These documents are added to the XISS data structures. Each document is assigned a document id (did). Index structures are organized as paged files for efficient disk IO. –When a query is performed the query processor interacts with XISS in order to obtain the information required for the query.

15 XISS - cont. XISS consists of 5 components: –Name Index –Value Table –Element Index –Attribute Index –Structure Index

16 Name Index and Value Table Objective: minimizing the storage and computation overhead by eliminating replicated strings and string comparisons. Name Index - mapping distinct name strings into unique name identifiers (nid). Value Table - mapping distinct value strings (i.e. attribute value and text value) into unique value identifiers (vid). Both implemented as a B + -tree.

17 The Element Index Objective: quickly finding all elements with the same name string. Structure:

18 The Element Index – cont. Structure: –B + -tree using nid as a key. –Leaf nodes: pointers to a set of records for elements (or attributes) having an identical name string, grouped by the document they belong to. –Element Record = {, Depth, Parent ID} where Depth is the depth of the element in the XML tree. –Element Records are ordered by.

19 The Attribute Index Objective: quickly finding all elements with the same name string. Structure: –Same structure as the Element Index except that the record in attribute index has a value identifier vid which is a key used to obtain the attribute from the value table.

20 The Structure Index Objectives: –Finding the parent element and child elements (or attributes) for a given element. –Finding the parent element for a given attribute. Structure:

21 The Structure Index – cont. Structure: –B + -tree using document identifier (did) as a key. –Leaf nodes: linear arrays with records for all elements and attributes from an XML document. –Each record: {nid,, Parent order, Child order, Sibling order, Attribute order}. –Records are ordered by order value.

22 Querying Method Decomposing path expressions into simple path expressions. Applying algorithms on simple path expressions and their intermediate results.

23 Decomposition of Path Expressions The main idea: –A complex path expression is decomposed into several simple path expressions. –Each simple path expression produces an intermediate result that can be used in the subsequent stage of processing. –The results of the simple path expressions are than combined or joined together to obtain the final result of the given query.

24 Basic Subexpressions - Example Decomposition of (E 1 /E 2 ) * / E 3 / ((E 4 | (E 5 /_ * /E 6 )): (1) Single Element/Attribute (2) Element-Attribute (3) Element-Element (4) Kleene Closure (5) Union / /_ * / *| [ ]/ / (4) (2) (3) (5) (3) (1)

25 Basic Subexpressions 5 basic subexpressions: (1) A subexpression with a single element or a single attribute. (2)A subexpression with an element and an attribute. e.g. = “Tree Frogs”] (3)A subexpression with two elements e.g. chapter/_*/figure where ‘_’ denotes any kind of node.

26 Basic Subexpressions - cont. 5 basic subexpressions - cont.: (4)A subexpression that is a Kleene closure ( +,*) of another subexpression. (5)A subexpression that is a union of two other subexpressions.

27 3 Algorithms 3 Algorithms: –EA-Join: Element and Attribute Join. –EE-Join: Element and Element Join –Kleene Closure

28 EA-Join: Element and Attribute Join Input: {E 1,…,E m }: E i is a set of elements having a common document identifier ( did ); {A 1,…,A n }: A j is a set of elements having a common document identifier ( did ); Output: A set of (e,a) pairs such that the element e is the parent of the attribute a.

29 EA-Join: Element and Attribute Join The Algorithm: // Sort-merge {E i } and {A j } by did. (1)foreach E i and A j with the same did do: // Sort-merge E i and A j by // PARENT-CHILD relationship (2)foreach e  E i and a  A j do (3)if (e is a parent of a) then output (e,a) end

30 EA-Join – Example Consider the XML document: And the query: Ele Att

31 Sort-merging “Ele”s and “Att”s by parent-child relation ship will give us the list:,,, Finding the elements “Ele”s with a child attribute “Att” with a value “A1” from the accepted list is easy using the information in the Element Record. EA-Join – Querying Ele Att

32 EA-Join – Comments Only a two-stage sort-merge operation without additional cost of sorting: –First merge: by did. –Second merge: by examining parent-child relationship. This merge is based on the order values of the element and attribute as defined by the numbering scheme. Attributes should be placed before their sibling elements in the order of the numbering scheme. –guarantees that elements and attributes with the same did can be merged in a single scan.

33 EE-Join: Element and Element Join Input: {E 1,…,E m } and {F 1,…,F m }: E i or F j is a set of elements having a common document identifier ( did ). Output: A set of (e,f) pairs such that element e is an ancestor of element f.

34 EE-Join: Element and Element Join The Algorithm: // Sort-merge {E i } and {F j } by did. (1)foreach E i and F j with the same did do: // Sort-merge E i and F j by the // ANCESTOR-DESCENDANT relationship. (2)foreach e  E i and f  F j do (3)if (e is an ancestor of f) then output (e,f); end

35 EE-Join – Comments Only two-stage sort-merge operation without the additional cost of sorting: –First merge: by did. –Second merge: by examining parent-child relationship. The sets of elements with a matching did cannot be merged in a single scan.

36 Kleene Closure Input: {E 1,…,E m }, where E i is a group of elements from an XML document. Output: A Kleene closure of {E 1,…,E m }.

37 The Algorithm: (1)Set i  1; (2)Set K i C  {E 1,…,E m }; (3)repeat (4)set i  i + 1; (5) set K i C  EE-Join(K i-1 C, K 1 C ); until (K i C is empty); (6) output the union of K 1 C,K 2 C,…, K i C ; Kleene Closure

38 Performance Experiments EE-Join: Results: –Real World: an order of magnitude faster. –Synthetic Data: 6 to 10 times faster.

39 Performance Experiments EA-Join: Results: –Compared to Top-Down: a better performance. –Compared to Bottom-Up: no winner - close results.

40 Performance Results - Conclusions The proposed algorithms can achieve performance improvement over the conventional methods (top-down and bottom- up tree traversals) by up to an order of magnitude.