XLIVE: INTEGRATING SOURCES WITH XQUERY

An Efﬁcient XQuery Evaluation Prototype with TGV

Nicolas Travers

PRiSM Laboratory, University of Versailles, 45 av des Etats-Unis, Versailles, France

Tuyˆet-Trˆam Dang-Ngoc

ETIS Laboratory, University of Cergy-Pontoise, 2 rue A. Chauvin, Pontoise, France

Tianxiao Liu

ETIS Laboratory, University of Cergy-Pontoise, 2 rue A. Chauvin, Pontoise, France and XCalia SA, Paris, France

Keywords:

XQuery evaluation, Extensible Optimization.

Abstract:

This paper presents an efﬁcient evaluation of XQuery in a heterogeneous distributed system.

XQuery(W3C, 2005) is a rich and so a complex language. Its syntax allows us to express a large scale of

queries over XML documents. We have extended (Chen et al., 2003) proposal to rewrite XQuery expressions

in ”canonical XQuery” in order to support the full XQuery speciﬁcation.

The XQuery expressiveness makes difﬁculties to obtain an exclusive internal representation within a system.

Models based on Tree Patterns have been proposed, and we have extended the tree pattern model to a model

called TGV that (a) integrates the whole functionalities of XQuery (b) uses an intuitive representation that

provides a global visualization of the request in a mediation context and (c) provides a support for optimization

and for cost information.

Our paper is based on the XLive mediation system. XLive integrates sources in a uniform view. It is a running

research vehicle designed at PRiSM Laboratory for assessing the integration system at every stage of the

process starting from sources extraction to the user interface and is already used in several projects.

This paper presents XLive, an efﬁcient XQuery

processing in a heterogeneous distributed system.

1 XLIVE ARCHITECTURE

The XLive prototype is designed to be a light media-

tion system with high modularity and extension capa-

bilities. It is a running research vehicle designed for

assessing the integration system at every stage of the

process starting from sources extraction to the user

interface, including query parsing and modeling, op-

timization and evaluation, and also benchmarking.

As most mediation systems (Wiederhold, 1992),

XLive is composed of three layers: Presentation, In-

tegration and Sources.

Figure 1 shows the integration system. On the

Source Layer, there are multiple heterogeneous data

sources (relational/object/XML data sources, web-

services, ﬁles, etc.) These sources are queryable by

XQuery XML

Wrapper

Source

Wrapper

Source

Wrapper

Source

XQuery Evaluator

Information

Wrapper

Manager

Console Benchmark Web Services

XLive public API

Query

Language

Data

Results

Presentation

Source

Integration

Figure 1: XLive Architecture.

the XLive system via wrappers. The Wrapper is a

component for accessing a speciﬁc Source for query-

ing and retrieving result. As sources have speciﬁc ac-

378

Travers N., Dang-Ngoc T. and Liu T. (2007).

XLIVE: INTEGRATING SOURCES WITH XQUERY - An Efﬁcient XQuery Evaluation Prototype with TGV.

In Proceedings of the Third International Conference on Web Information Systems and Technologies - Internet Technology, pages 378-381

DOI: 10.5220/0001275203780381

 SciTePress

cess methods, the role of the wrapper is to translate

the wrapper speciﬁcity to a common access method.

The Integration Layer is the heart of the media-

tion system as it process XQuery requests according

to sources and send by the result in XML form. The

Wrapper Information Manager is for integrating in-

formation about wrapped sources. It provides to the

mediator sources metadata, capabilities, and costs sta-

tistics on source data. The XQuery Evaluator parses

the XQuery query, make logical plans then physical

plans by using optimization rules, choose the best ex-

ecution plan, and then evaluates the execution plan

by querying relevant sources and merging results The

XLive system is provide a public API that process an

XQuery query and evaluates it on sources, and then

return the result as an XML document.

Some clients using the XLive API have been im-

plemented on the Presentation Layer: The Console

is a graphical interface for managing and querying

XLive. The Web service provides services for man-

aging/querying XLive from the Web. And ﬁnally, the

Benchmark is designed to test the XLive mediator and

other sources in order to make comparisons.

More details on the XLive architecture can be

found on (Dang-Ngoc and Gardarin, 2003; Dang-

Ngoc et al., 2005). The TGV model used for XQuery

evaluation has been described in (Travers and Dang-

Ngoc, 2006a; Travers et al., 2007). And a paper de-

scribing the TGV annotation method and extensible

optimization rules has been submitted to this confer-

ence in the paper session.

2 XQUERY EVALUATION

XQuery is a rich and complex language. Its power-

ful expression capabilities provide a large range of

queries over XML documents. However the richness

of the language makes the evaluation of XQueries in-

efﬁcient in several cases.

Figure 2 describes the evaluation process: (1)

XQuery is canonizedinto a canonicalform of XQuery

(2) then the canonized XQuery is modeled in the in-

ternal structure TGV which can (3) be restructured

into equivalent structures using equivalence rules. (4)

Then the TGV is annotated with information for eval-

uation such as the data sources location, cost models

information, sources functional capabilities, etc. The

optimal annotated TGV is then selected and (5) the

logical TGV is transformed into an execution plan us-

ing a physical algebra. We have chosen the XAlgebra

(Dang-Ngoc and Gardarin, 2003), an extension of the

relational algebra to XML. (6) Finally, the execution

plan is evaluated and produces an XML result.

logical plan

physical plan

XML

XQuery

XAlgebra

Annotated

TGV

Canonized

XQuery

TGV

transformation

into a physical

algebra

optimization

(physical rules)

(logical rules)

optimization

Preparing the execution plan

annotation

modelizationcanonization

evaluation

Figure 2: XQuery evaluation process.

Table 1: An XQuery query example.

<bib>{

for $b in doc("http://bstore1.example.com/bib.xml")//book

where count ($b/author) > 0

return

<book>

{$b/title}

{$b/author[position() <= 2]}

{if (count ($b/author) > 2) then <et-al/> else ()}

</book>

}</bib>

The whole process is implemented in the XLive

(Dang-Ngoc et al., 2005) system and validates all use-

cases deﬁned by the W3C that do not imply strong

typing consideration (our system recognize 8 of the 9

categories of XQuery).

Table 1 givesan XQuery query example supported

by our XQuery processor. It is the query XMP-6 de-

ﬁned in the W3C use-cases document. This query ex-

presses the following request: ”For each book that

has at least one author, list the title and ﬁrst two au-

thors, and an empty ”et-al” element if the book has

additional authors.”

2.1 XQuery Canonization

XQuery deﬁnes complex operations such as ag-

gregation, ordering, nesting/unnesting, document

construction, conditional cases (if-then-else),

and XPath ﬁlter. To handle such functionali-

ties, a canonical form using simple sequences of

FOR...LET...WHERE...ORDER-BY...RETURN

(FLWOR) expressions can be used (the demonstra-

tion begin in (Chen et al., 2003) and achieved in

(Travers and Dang-Ngoc, 2006b)).

In order to optimize the evaluation of such queries,

we propose to structure the language declaration by

canonizing it using some transformation rules. These

transformation rules keep the semantic of queries and

make them more convenient to manipulate.

XLIVE: INTEGRATING SOURCES WITH XQUERY - An Efficient XQuery Evaluation Prototype with TGV

379

Table 2: Canonized XQuery query.

let $r_1 :=

for $b in doc("http://bstore1.example

.com/bib.xml")//book

let $r_2 := for $a in $b/author

where position ($a) <= 2

return $a

let $agg_1 := count ($b/author)

where $agg_1 > 0

return

<book>

{$b/title}

{$r_2}

{if ($agg_1 > 2) then <et-al/> else ()}

</book>

return

Table 2 gives the canonized XQuery form of query

shown of the example presented in table 1. The fol-

lowing canonization rules have been used:

• Each FLWR-type queries nested in the return

clause is redeﬁned in a let clause and is assigned

to a variable. In the return clause, the nested

queries are replaced by their assigned variable.

• Each aggregate function is redeﬁned in a let

clause and is assigned to a variable that replaces

the aggregate function in the main expression.

• Each quantiﬁer functions (every and some) is re-

deﬁned in a let clause and is assigned to a variable

that replaces the quantiﬁer functions in the main

expression.

• Nested expressions in conditional expressions

If/Then/Else are declared in let clauses with new

variables. The conditional expression is rewritten

using those variables.

2.2 TGV

XQuery modeling is a difﬁcult goal since it pro-

vides a large set of functionalities. Its canonization

in a canonical form helps us deﬁning a more sim-

ple model. Moreover, mediation purpose directs our

reﬂection for modeling in speciﬁc Tree Patterns in

which all XQuery operations are represented in a sin-

gle expressive form. Each set of trees deﬁned in can-

onized queries generates a particular representation.

Tree Pattern Queries (Sihem et al., 2002; Jagadish

et al., 2001) are now well admitted for modeling parts

of XML Queries. Works as GTP (Chen et al., 2003)

use the Tree Pattern Query as a basis to model a part

of the XQuery speciﬁcation.

We propose the TGV (Tree Graph View) model

which is implemented in the XLive mediator to

process XQuery. The TGV model (Travers and Dang-

Ngoc, 2006a; Travers et al., 2007) extends the Tree

Pattern representation in order to make it intuitive and

full non-typed XQuery support. XQuery queries are

modeled in TGV, then optimized, translated to an ex-

ecution plan and evaluated.

The TGV model allows us to cover 8 of 9 use

cases of the W3C (unless strong typing). In fact, spe-

ciﬁc XQuery characteristics have been integrated into

the TGV model by canonization or speciﬁc elements.

Figure 3: TGV.

The TGV proposed in ﬁgure 3 illustrates the can-

onized query (table 2). We can identify tree pat-

terns for source document, return document and in-

termediate structures. On tree patterns are represented

by labels, constraints and ascendant-descendant links

(simple line for descendant axis, double link other-

wise), and a variable is bound. Between trees, hyper-

links represent aggregation, join, projection or condi-

tion. More description on TGV are given in (Travers

and Dang-Ngoc, 2006a; Travers, 2006).

3 THE XLIVE SYSTEM

The whole XLive architecture with TGV support and

full untyped-XQuery implementation has been re-

leased as an Open-Source software

In the ﬁgure 4, we show a XLive screenshot where

we see

 data sources management

 global schema

management

 an XQuery query

 the associated

logical TGV

 the associated XAlgebra execution

plan and

 the XML result document.

A benchmark (Dragan and Gardarin, 2005) has

been created in the context of distributed semi-

structured model on heterogeneous sources. It has

The XLive system is downloadable at

http://www.prism.uvsq.fr/˜ntravers/xlive/

WEBIST 2007 - International Conference on Web Information Systems and Technologies

380

Figure 4: XLive execution.

been speciﬁed and applied to XLive. It performs com-

parisons between queries evaluated on single sources

and queries evaluated through the mediator. Medi-

ator performance is shown in Figure 5 for a set of

representative queries (Selection, projection, join, re-

construction). We use two kinds of XML data in our

scenarios, data oriented and structure oriented. XLive

stores those data in different systems: native XML

repositories (XHive, Xyleme), relational systems (Or-

acle, MySQL) and web sources (Google, Amazon).

There is an overhead due to the mediation process

compared to single source evaluation. The execution

is 1.4 to 2 times slower.

Figure 5: Benchmark.

ACKNOWLEDGEMENTS

The XLive is supported by the ACI Semweb project.

The cost model is supported by the ANR PADAWAN

project and XCalia S.A.

REFERENCES

Chen, Z., Jagadish, H., Laksmanan, L. V., and Paparizos, S.

(2003). From Tree Patterns to Generalized Tree Pat-

terns: On efﬁcient Evaluation of XQuery. In VLDB,

pages 237–248, Germany.

Dang-Ngoc, T.-T. and Gardarin, G. (2003). Federating

Heterogeneous Data Sources with XML. In Proc. of

IASTED IKS Conf.

Dang-Ngoc, T.-T., Jamard, C., and Travers, N. (2005).

XLive: An XML Light Integration Virtual Engine. In

Proc. of BDA.

Dragan, F. and Gardarin, G. (2005). Benchmarking an XML

Mediator. In ICEIS, Miami USA.

Jagadish, H., Lakshmanan, L. V., Srivastava, D., and

Thompson, K. (2001). TAX: A Tree Algebra for

XML. In DBPL, pages 149–164.

Sihem, A.-Y., SungRan, C., Laks, L. V. S., and Divesh,

S. (2002). Tree Pattern Query Minimization. VLDB

Journal, 11(4)::315–331.

Travers, N. (2006). Extensible Optimization in an XML Me-

diator. PhD thesis, University of Versailles.

Travers, N. and Dang-Ngoc, T.-T. (2006a). Tree graph view

(tgv). Technical report, PRiSM Laboratory.

Travers, N. and Dang-Ngoc, T.-T. (2006b). Xquery canon-

ization. Technical report, PRiSM Laboratory.

Travers, N., Dang-Ngoc, T.-T., and Liu, T. (2007). Tgv:

an efﬁcient model for xquery evaluation within an in-

teroperable system. International Journal of Interop-

erability in Business Information Systems (IBIS), 2.

ISSN: 1862-6378.

W3C (2005). An XML Query Language (XQuery 1.0).

Wiederhold, G. (1992). Mediators in the Architecture of

Future Information Systems. Computer, 25(3):38–49.

XLIVE: INTEGRATING SOURCES WITH XQUERY - An Efficient XQuery Evaluation Prototype with TGV

381