International Lattice Data Grid

1
International Lattice Data Grid

• Richard Kenway
• University of Edinburgh

2
lattice QCD

• computer simulation of the strong force
• necessary to seek new physics which might hold
  clues to the Theory of Everything
• national projects using Tflops computers
  generating TByte data sets
• riding exponential technology curves
• towards petascale

3
virtual organisation

• grand vision
• shared primary data
• shared codes
• shared machines
• reality
• Australia, Europe, Japan and US, led by UKQCD
• focus on data sharing
• grid of grids
• connect existing data repositories using web
  services
• provide a single public web interface
• metadata
• QCDML and data binding
• draft standard XML schema for primary data

4
UK QCDgrid
QCDML
Tier 3
Edinburgh
RAL
Linux PC RAID
Liverpool
QCDOC
Tier 2
Swansea
Tier 1
ILDG via SRM
5
QCDgrid operational model

• control thread
• checks disk space available on each node
• maintains geographically separated copies of each
  file
• moves each file closest to its point of use
• can be transferred to another node
• replica catalogue
• tracks number and location of file instances

6
QCDML

• minimal set of XML elements for exchanging
  primary lattice data
• physics parameters
• QCDML glossary documents on the web for
  algorithm, implementation and physics details
• extensible to other data

schema
instance document
ltplaquette_gauge_actiongt ltplaquette_operator
gt ltfieldgtlink_gluonlt/fieldgt
ltcouplinggt ltbetagt 5.2 lt/betagt
lt/couplinggt lt/plaquette_operatorgt
ltplaquette_gauge_actiongt
7
data format strategy
C library (provided by contributor) to read data
into standard format
convert it to user's format if necessary
OR
BinX description of format B

• BinX is an XML schema to describe binary file
  format developed by the eDIKT project
• allows data binding

8
a query using the metadata browser
view the XML
click get data
9
grid of grids

• QCDgrid and JLab are running SRM v.1 as an
  interface
• SOAP based XML protocol
• JLab JAVA browser as a GUI for web services
• today
• can list QCDgrid and JLab LFNs as an aggregation
  of the replica catalogues
• next steps
• include Japans CP-PACS data
• establish trust network of Certification
  Authorities or ILDG CA
• increase functionality of the SRM
• interact with the metadata