Dr Chris Maynard Application Consultant, EPCC [email protected]@ed.ac.uk +44 131 650 5077 Muttering...

Dr Chris MaynardApplication Consultant, EPCC

[email protected]+44 131 650 5077

Muttering about metadata

Report from the Metadata work group

Review of QCDml

metadata

• Meta- Greek among, with, after, from

• Data Latin information

• Literally data about data

• Data– Gauge configuration – Ensemble of gauge configurations

• Metadata (MD)– How was data created

– Format, machine, code, algorithm, physics

210-12/03/2009 XML at light quarks @css

Why do we need metadata?

• Extreme example no metadata– Cfgs have random string names with no directory structure for

different ensembles– Impossible to use

• Organise files– Into directories for ensembles

– Give cfgs names with markov chain position

• Construct a scheme for the metadata– Rules for describing the data


A basic scheme

• Use “meaningful” filenames– Metadata is encoded into the names of the files and

directories– Can have some structure or hierarchy

– But not completely flexible– Example with three fields. What ordering?

–gauge-action/volume/quark-mass–gauge-action/quark-mass/volume

– What happens for 2+1 flavours– Not extensible


D52C202K3500U010010_LL3450X_FL3400X_CMesonT00T31

Old UKQCD meson correlator filename– What does X stand for?

Wilson, Rotated, Clover– Many different clover. Scheme broken– X means none of the above!

D52C202K3500U010010_LL3450X_FL3400X_CMesonT00T31

Dynamical cSW=2.02 ≈ 2.0171

NP determined – no information


A broken scheme

A better scheme

• Recreate data from MD– This is an very important requirement

– Know what the data is – Data provenance

• Combination of IO parameters and code– Implicit assumptions are recorded! :^):^)– Version n cannot read version m :^(:^(– Code X cannot process MD from Code Y :^(:^(

• Can we construct a general scheme?– Recreate data from MD?


Extensible schemes

• LQCD metadata is hierarchical– Rich structure– Metadata scheme has to reflect this

– Extensible

• New metadata requires a new scheme– In extensible scheme

– Old scheme is included in new one– Old metadata fits in new scheme– No need to refactor existing documents


Markup language

• Combines text and information about text– Presentational

– text format (e.g. This slide) WISIWIG– Procedural

– presentation of text, not WISIWIG– Tex, postscript

– Descriptive or semantic– Labels fragments of text– No presentation or other interpretation mandated– SGML, XML, VML

HTML has both procedural and descriptive elements


• XML eXtensible Markup Language• www.w3.org/XML

– XML is for structuring data

– XML looks a bit like HTML

– XML is text, but isn't meant to be read

– XML is verbose by design

– XML is a family of technologies

– XML is license-free, platform-independent and well-supported

– <beer>happiness</beer>


XML Web standard

XML II• Semantic, eXtensible Markup language

• XML was designed to carry data, not to display data– Cf. with HTML, designed for displaying data.– Incompatible applications can exchange data wrapped in xml

• XML is just plain text

• User defined tags allow structure to be developed– Lattice QCD metadata is structured

• XML does not DO anything– You need an application for this

• XML schema– Defines a set of rules for the XML document

10-12/03/2009 XML at light quarks @css 10

Well formed XML


XML schema

• What is XML schema?– Collection of rules for XML documents

– Other schema languages, DTD, Relax NG, Schematron

– An XML schema is itself an XML document

• Why do we need an XML schema?– Computers can read and understand XML IDs– <length>16</length>– Meaning of length is context dependent– Applications can know types, parse and processes

XML data– Could just be an XSLT style sheet to transform XML

in HTML and render a web page e.g. LDG web-client


A word of caution

• XML is not magic– XML is not a solution– It is a useful tool– Not the only tool– We still have to use the tool

• Ideally produce metadata from code– What metadata?

– What is standard/useful/implicit– Application has to do something with it


Metacrap

• Not all aspects/connotations of metadata are good

• Metacrap: Putting the torch to seven straw-men of

the meta-utopia

• http://www.well.com/~doctorow/metacrap.htm

• Amusing and valid critique of some MD ideas– But not all are relevant to this project


International lattice data grid (ILDG)

• Lattice data is very expensive to generate

• Many groups now make data available– MILC, UKQCD, RBC, JLQCD, Adelaide group– Many others share data

• ILDG is representative of whole lattice community– MD working group (CMM is UK rep)– Middleware working group (MGB, RO [epcc])


Introducing QCDml

• XML schemata for gauge configuration MD

• Developed and maintained by MDWG– Design by committee – always a good idea

• Basic concept– MD describing an ensemble– MD describing a configuration belonging an ensemble


QCDml Ensemble


Physics


Fermion action inheritance


Example QCDml Ensemble

• 1 Name of schema (URI)• 2 Using W3.org XML schema (URI)• 3 Schema location for

– a) named schema (URI) – b) Location of schema (URL)

• 4 Name of Ensemble (URI)

20XML at light quarks @css

Example quark action


QCDml Config


Markov Step


Name hierarchy

Unique name in for ensemble in ILDG namespace

<markovChainURI/> Ensemble

<markovChainURI/> Config

<dataLFN/>

Replica catalogue

Actual file instances Multiple copies


Algorithm

• General scheme too complex• Algorithmic MD can belong to ensemble or

config.• Either name value pairs• Or import another schema

– Lives in separate namespace


Namespaces

Allow another namespace to be imported here

Application processing QCDml can ignore this namespace

Can include all metadata into XML IDs

Local applications can be alg schema aware, but ignore non-local ones


Alg Namespace example

Imported namespace has it’s own schemaImported schema is not in default namespace, so has a prefixAll elements belonging to this namespace use this prefix


Data format

• ILDG specified format– All gauge configurations in the same format

• Based on NERSC data layout

• SciDAC LIME records

• UKQCD perspective– CPS cannot do this :^(:^(– Chroma can :^):^)– qdp++ tools exist for conversion etc


How to generate QCDml

• Ensemble MD requires a human– Use schema aware tool (cmm uses XMLspy)– Take existing XML ID and hack– Not that hard, only once per ensemble

• Config XML– Post-processing on QCDOC– Example of DWF data


DWF data

• CPS on QCDOC writes– Data in NERSC format– VML files containing MD

– Parameters of objects– Effective check-pointing

– Data stored with meaningful path- and filenames– Includes binary and source code version

• Satisfies important constraint– Recreate data from Metadata :^):^)


From one scheme to another

• Series of scripts and utilities can do conversion– qdp++ – cmm – /host/cmaynard/tools/scripts

• Utilities compiled for qcdocx– Conversion and submit to grid


Scripts

• Data conversion and

XML chunks are built

by scripts

• makeQCDmlConfig.sh glues

XML together

• dataSet.sh contains

dataset specifications– Plus sundries


Chris Maynard ILDG 13 December 5 2008 36

Problems with XML

• Lattice QCD (meta)data is really mathematics– XML is not really ideal for storing this data

• QCDml has defined common names for <action/> etc– Even WilsonAction has more than one common usage

– Kappa versus mass

• Algorithm metadata is too complex for common names– Not really defined in the metadata– Unstructured parameter values included

• This is OK because – an ensemble is defined by the action – not the algorithm used to generate it

• Extending to propagators and correlators is hard for the

same reason as defining the algorithm


We need an application• XML does not DO anything• For it to be useful we need to do something with it!

– What do we want to do with it?– Is QCDml good for this purpose

– QCDml design focused on searching the metadata catalogue– This was probably a good idea!

• Xpath used to query XML databases– Basic tools/APIs exist for constructing queries

– Cf. UKQCD DiGS GUI browser, LDG web-client and JLDG faceted navigation application

• Metadata capture– How do we create XML IDs?

– Does any application actually write QCDml?– UKQCD does post-processing

• Data provenance– Does QCDml provide this?

Hard-work


What next?

• QCDml seems to work OK– How much is it being used?

• We don’t have many applications that DO something

with it CMM’s Questions for ILDG

– What do we want to do with metadata?

– Do we have the right sort of metadata for this?

– What tools or applications do we need?

• Someone then has to build them• if we don’t ask, we don’t get!

Can we review QCDml usage to define what tools we need?

Work flow tools

• Graphical tools for linking work together– components could be …– Machine job submission tasks– The actual MC code– data logistics

• Now used by many areas of science, – particle physics experiments– chemistry

• Examples– Unicore– “my experiment”


Topics for Discussion

• Technical– tools

– Metadata capture– Data conversion

– Use of QCDml– Data Curation, – Data provenance

• Sociological – Encouraging other groups to join – ILDG paper– Funding


Finally


Dr Chris Maynard Application Consultant, EPCC [email protected]@ed.ac.uk +44 131 650 5077 Muttering...

Documents

Transcript of Dr Chris Maynard Application Consultant, EPCC [email protected]@ed.ac.uk +44 131 650 5077 Muttering...