Web Site Management Based on Declarative Specifications

1
Web Site Management Based on Declarative
Specifications

• Alon Levy
• University of Washington
• Joint work with
• Strudel Dana Florescu (INRIA), Mary Fernandez,
  Dan Suciu (ATT), Khaled Yagoub (INRIA)
• Tiramisu Corin Anderson and Dan Weld (UW)

2
Problem Building Web sites

• Building Web sites involves three tasks
• Selecting and managing the sites content
• Organizing the sites structure (pages and
  links)
• Designing the graphical presentation of pages.
• In current tools, these tasks are (mostly)
  interdependent.
• Strudels key ideas
• Separate the three tasks.
• Manage content and structure declaratively.

3
Content Management and Graphical Presentation

• Content may be derived from multiple sources
• Databases relational, object-oriented
• Semi-structured sources (XML, Word, Excel,
  bibtex).
• Classical data integration problem!
• (see Tsimmis, Garlic, Information Manifold,
  Tukwila)
• Graphical presentation
• Need to integrate with tools that create
  animations, images, Java applets.
• Create sets of similar HTML pages using
  templates.

4
Web-Site Structure

• The structure includes
• Set of pages and contents of each page, and
• Links between the pages.

5
Current practice

• Current tools separate only content management
  from presentation
• Content managed by database
• Embed queries in HTML templates
• Simple tools to view and modify structure at the
  extensional
• level.
• WYSIWYG tools for managing presentation.
• But they still cannot
• explicitly manage site's global structure, or
• flexibly choose content-management system
• As a result its hard to
• modify the structure of a web-site, build
  multiple versions for
• different classes of users, enforce integrity
  constraints.

6
Talk Outline

• Problem definition
• Strudel architecture
• Advantages of declarative specifications
• Specifying and verifying integrity constraints.
• Automatic generation of run-time plans for
  managing data-intensive web sites.
• Tiramisu
• Separating the design tool from the
  implementation.
• Using a collection of tools to build a site.

7
Strudel Evolution
Strudel (Nov. 96)ATT
Strudel ATT Release
Strudel-R (INRIA)
http//www.research.att.com/sw/tools/strudel
Tiramisu (Sept. 98) (U. Washington)
8

Strudel Architecture and System
9
Strudel

• Features
• Integrates content from multiple sources.
• High-level declarative language for managing
  sites structure (StruQL).
• Advantages
• Derives multiple sites from the same data.
• Supports easy restructuring and modification.
• Provides platform for
• Enforcing integrity constraints
• Designing policies for efficient run-time
  management of sites.

10
Strudel Architecture
11
Data Model

• Strudel is based on a semi-structured data model
  
• labeled directed graphs.
• nodes in the graph represent objects,
• labels on arcs represent attribute names,
• named collections.
• Why semi-structured data?
• raw data is often semi-structured (and I dont
  mean that its
• embedded in HTML)
• convenient for data integration (a la TSIMMIS)
• web-sites are ultimately graphs.

12
The StruQL Query Language

• A StruQL query is a function from a set of input
  graphs to an
• output graph.
• A StruQL expression contains two parts
• A query component, and
• A restructuring component.
• Formally
• INPUT graph names
• WHERE conjunction of regular path expression
  atoms
• CREATE name the nodes in the output graph using
  Skolem functions
• LINK specify the links in the resulting
  graph.
• StruQL evolved into XML-QL, (see WWW8 Conference)

13
Example Raw Data

• Article 1
• Date 8/1/97
• Title Clinton announces new
• Priority Headline
• Category USA News
• Images im1.gif, im.gif
• Text President Clinton announced
• Related article article2

• Article 2
• Date 8/2/97
• Title FDA approves new cure for
• Priority Top Story
• Category Health
• Video vid1.avi
• Text The Federal Drug Administration

14
CNN Web-site Query (part 1)
Input graph of articles INPUT CNN-ARTICLES Create
web page for each article WHERE Articles(a),
note arc variable
l art - l - t, l in "Title",
"Abstract", "Date", "Text",
"Image", "Topimage", "RelatedSite", a -
"Category" - c CREATE ArticlePage(a) LINK
ArticlePage(a) - l - t WHERE a -
"RelatedArticle" - r LINK ArticlePage(a) -
"RelatedArticle" - ArticlePage(r)
15
CNN Site Schema
RootPage()
a- priority- headline
a- category-c
CategoryEntry(c)
RootPageEntry(a)
Data(t)- a - l -t l in title, top-image
CategoryPage(c)
a -category-c
ArticlePage(a)
Data(t) a - l - t, l in "Title",
"Abstract",
16
CNN Web-site Query (part 2)
CREATE RootPage WHERE a - "Priority" -
"headline", l in "Title",
"Date", "Topimage" CREATE RootEntry(a)
LINK RootPage - "HeadlineStory" -
RootEntry(a), Link each headline story to its
title, date, top image and full article
RootEntry(a) - "FullStory" - ArticlePage(a),
RootEntry(a) - l - t
17
HTML Templates
EMBED , EMBED related-article ORDERdescend KEYdate _at_a LINKtitle

18
CNN Sports Query
INPUT CNN WHERE TopCategory(c), c -
"CategoryName" - cn,
cn"Sports", c - "SubTopic" - top,
Articles(a), a - l - t, l in
"Title", "Abstract", "Date", "Text", "Image",
"Topimage", "RelatedSite", a
- "Category" - c, ctop CREATE
ArticlePage(a) LINK ArticlePage(a) - l - t
19
StruQL Details

• Regular path expressions are constructed by a
  grammar
• R _
• Atoms in the WHERE clause are of the form X - R
  - Y or C(X)
• The LINK clause includes atoms of the form
• LINK f(X) -- new link -- g(X) or
• LINK f(X) -- L -- g(X)
• Queries can be nested, inheriting the WHERE
  clauses of
• their outer blocks.
• Note separation between querying part and
  restructuring part!

20
More on StruQL

• Bare bones language for semi-structured data
  includes the essential features.
• More expressive than Lorel or UnQL (e.g., can
  reverse graphs)
• Conceptually and in practice separation between
  query component and restructuring component is
  important.
• Containment is decidable for StruQL-WHERE
  (Florescu, Levy Suciu, PODS-98)

21

Advantages of Declarative Specifications
22
Enforcing Integrity Constraints

• We often want to verify some constraints on site
  structure
• all articles from the last two days are reachable
  from the root
• all paths to confidential data must go through an
  authentication node
• Good site design principles are summarized as
  integrity constraints Lohse, CACM, 98.
• When site specs are long, constraints are hard to
  enforce.
• Want to verify constraints intentionally.

23
Intentional IC Verification

• Formally, we want to check whether
• S(D) IC
• S is the site specification (e.g., StruQL Query)
• IC is a formula describing the constraint
• ? a, Article(a) date(a) today-2
• Root - -
  ArticlePage(a).
• for any instance D of the underlying data.
• Results
• Sound and complete algorithms for verification of
  a class of integrity constraints (path
  constraints).
• Algorithms will also propose corrections when
  ICs are violated.

24
Run-time Management of Sites

• When do we compute web pages?
• Static approach completely precompute site
• Doesnt work for large sites, forms, hard to
  update.
• Dynamic approach compute pages on request
• Users may wait, a lot of repeated computation,
  structure of the site is not exploited.
• Current tools use one of the extremes, or specify
  policy per collection of pages.
• The specification is implicit in code.
• Our goal use site specification to automatically
  find optimal strategy.

25
Possible Run-time Optimizations

• View materialization
• Function caching
• when web sites represent hierarchically
  structured data, successive queries in the site
  differ only in their projected attributes.
• Simplification under preconditions
• previous queries on the path may have already
  verified some conditions for current query.
• Lookahead computation
• often it is possible with little cost to compute
  the data necessary for subsequent pages.

26
Problem Statement

• Given
• site specification
• knowledge about browsing patterns
• cost function
• Produce
• Operational plan
• operational schema a set of queries to
  compute on a given page request.
• Results (in Strudel-R) framework
• Performance study of the optimizations.
• Algorithm for generating operational plans.
• Identification of many open problems.

27

Strudel Experience -- Tiramisu
28
Experiences with Strudel(except for the lousy
GUI)

• Integrating data from multiple sources when
  building a Web site
• is a prime concern. Sources are
  semi-structured!
• Declarative specification of site structure is
  very important
• because
• site creation is a highly iterative process
• site owners often need redesign after
  experience from
• deployment
• we often generate multiple versions of sites
  from the
• same data.
• Design of web-sites is done in a top-down
  fashion.
• Strudel cant be the all encompassing web-site
  management tool.

29
Tiramisu the Second Generation

• Strudel and its siblings (Araneus, YAT, WebOQL,
  WIRM) force the design and implementation of the
  site to be done in the same tool.
• Furthermore, there will always be tools that are
  specialized for specific tasks.
• Tiramisu
• Separate design phase from implementation.
• Allow the implementation to be done by a set of
  cooperating tools.

30
Tiramisu Architecture
mediator
data source
E/R style diagram of site (site schema)
data source
web site
Implementation manager
data source
wrapper
wrapper
wrapper
Tool (ASP)
Tool (FrontPage)
Tool (Strudel)
31
Screenshot of a TERD
32
Conclusions

• Web-site management is an important area for
  Database research.
• First-generation systems (Strudel, Araneus, YAT,
  WebOQL) offer important advantages
• Easy modification, creation of multiple versions
• enforcing constraints, run-time management
• Second generation (Tiramisu)
• Emphasize design phase of site
• Implement with a collection of cooperating tools.