tt dafea

1
DBGlobe IST-2001-32645
WP4 Querying and Information Discovery INRIA
(Serge Abiteboul)
Proactive initiative on Global Computing (GC)
Future and Emerging Technologies (FET)
The roots of innovation
2
Main choices

• Main technical issues for querying in the DbGlobe
  context
• Exchange of information between PMOs
• Management of distribution and replication
• provides support for proxy, caching
• Requirement use standards
• XML
• XML query languages (XPATH, Xquery, XOQL)
• Web services (SOAP)
• Proposed solution Active XML
• Discovery of information linked to discovery of
  services, see WP1

3
Outline for Active XML

• Active XML data integration and querying
  platform
• The AXML language
• AXML peers repository Web client Web server
• System architecture
• Distribution and replication Sigmod03a
• Motivation mobile devices with limited resources
• Contributions
• Controlling service invocations Sigmod03b
• Motivations
• A technique based on typing
• Conclusion

4
Active XML

• XML document with embedded calls to Web services
• Intensional information
• Dynamic information
• AXML peer (e.g., a PMO)
• Repository stores (A)XML data
• Client calls Web services
• Server support for queries over repository data
• Foundations to be further investigated
• Deductive databases and fixpoint logic
• Tree automata
• Rewriting systems

5
AXML Documents

• ltknownAuctions ID"peer10"gt
• ltcategory name"Toys"gt
• ltscgteBay.net/getOffers("Toys")lt/scgt
• ltauction id"1254" gt
• ltheldBygteBay.netlt/heldBygt
• ltitemgtPink pantherlt/itemgt
• lt/auctiongt
• ltscgtbabel.org/translate("Czech", "English",
• ltscgtcrystal.cz/getToys()lt/scgt)lt/scgt
• ltscgtpeer25/getAuctions(../_at_name)lt/scgt
• lt/categorygt ...
• ltsc frequency"once" gtgetMyAuctions()lt/scgt
• lt/knownAuctionsgt

• May contain calls to any SOAP Web service
• Are enriched by each service call's results
• The returned nodes are inserted as siblings of
  the corresponding ltscgt element
• Activation of calls and data lifespan are
  controlled
• frequency when is the service called ?
• validity how long is the retrieved data kept ?
• mode immediate or lazy ?
• Push and pull modes (see WP3)

6
AXML Services

• let closeAuction(a) be
• for b in a/bid
• where b/amount max(a/bid/amount)
• return
• ltsc mode"immediate" frequency"once"gt
• notifyWinner(b/who, a/aID, b/amount)
• lt/scgt
• ltstatusgtclosedlt/statusgt

• A simple, declarative way to create Wef services
• A service operation is specified as a query with
  parameters.
• It may query (local) AXML documents.
• It is made available on the Web using the SOAP
  protocol.

Basic AXML services are compatible with current
standards for Web services invocation.

• which allows for new, powerful features.
• Intentional parameters and results AXML
  documents (containing service calls) can be
  exchanged.
• Continuous services send back a stream of
  answers (SOAP messages) to the caller.

Used in AXML documents, AXML services are
powerful tools for data integration and querying.
7
AXML Architecture

• Technical environment
• SUNs Java SDK 1.4 (includes XML parser, XPath
  processor, XSLT engine)
• Apache Tomcat 4.0 servlet engine
• Apache Axis SOAP toolkit 1.0 beta 3
• X-OQL query processor, persistant DOM repository
• JSP-based user interface, using JSTL 1.0
  standard tag library

8
Example A number of PMOs engaged in auctions

• Each PMO proposes auctions
• Document myauctions.xml with the PMOs items and
  their current bids
• Services offered
• getLocalAuctions(),
• status(auctionId)
• Each PMO bids on auctions
• Document mybids.xml with the PMOs bids
• Service offered
• bid(PMO,auctionId, amount)
• bidUpTo(PMO, auctionId, increment, limit)

• Each PMO knows about other PMOs auctions
• Document allauctions.xml contains calls to other
  PMOs that transitively retrieve their known
  auctions.
• Service offered getAllAuctions()
• When an auction closes, the winner is notified.
• VLDB02 demo

9
Distribution and replication
10
Motivation

• PMOs have limited resources
• Storage space
• Computation power
• Network bandwidth
• Therefore, we would like to
• distribute the load among PMOs, by
• Using distributed services (AXML does it !)
• Distributing documents across several devices
• replicate documents and services, to allow for
  local computation.

11
Example

• A user has a PDA, and is in Colorado.
• We could let her call the ski portal services,
• We would rather replicate on her PDA the relevant
  data and services, so that they can be called
  locally.

• A national US ski portal
• Contains information about ski resorts, hotels.
• Provides services to query this information.
• Issues
• Data / services should be replicated ?
• How to adapt the replicated code ?
• The PDA has limited resources / capabilities,
  therefore
• We need a cost model to choose what to replicate
• Control the use of replication

12
Contributions

• We developed
• A data model for AXML with distribution and
  replication
• A location-aware extension of XQuery to handle
  distributed/replicated data and services.
• A cost model for the peer-to-peer context,
  measuring the observable performance of each
  peer
• A replication algorithm (based on the cost model)
  that determines
• What data/services need to be copied ?
• How to adapt the service code ?
• Implementation work is about to start.
• A paper describing the results, to appear in
  SIGMOD03

13
Controlling service invocation
14
Controlling service invocation

• Active XML peers call each others services,
  therefore they exchange AXML documents, that may
  contain service calls.

(A)XML documents

• When such a document is exchanged, the services
  calls it contains can be invoked
• by the sender (before sending the document)
• or the receiver (after receiving it)
• How do we control it?

SOAP service
SOAP client
15
Controlling service invocation why?

• For security reasons or capabilities, e.g.
• I dont trust this service/domain
• I dont have the right credentials to invoke it,
  or it costs money.
• I dont know how to deal with a service call,
  because I dont know Active XML !
• Decide where to perform some calls
• It may be better to have a proxy perform all the
  calls on behalf of a PDA
• Improve performances
• and many others.

16
How we do it With types !

• WSDL Web services description language (XML
  schema)
• We extended it to include service calls
• We provide algorithms to decide to call or not
  to call
• It finds which service calls should be invoked
• This is an inherently recursive process, since
  these calls may return data that contains service
  calls
• Foundations alternating tree automata techniques
• General problem still open progress
• A prototype has been implemented
• A paper describing the results, to appear in
  SIGMOD03

17
Conclusion

• Active XML Version 1 is out
• Extensions of AXML to fit DBGlobe requirements
• Design of distribution replication
  implementation has started
• Design of the control of call invocation has
  been implemented and is now being integrated
• Work on optimization has progressed
• Very complex issue
• Involves many facets of optimization of
  distributed databases novel issues
• More work needed
• AXML PMO on a small device
• Implementation for coming year

18
Thank you.