A Data Analysis Framework for the Neutron Community

1

(Distributed Data Analysis for Neutron
Scattering Experiments)
A Data Analysis Framework for the Neutron
Community
Michael M. McKerns Materials Science and
Applied Physics Center for Advanced Computing
Research California Institute of Technology
2
Serving a Growing Community

• With the availability of OPAL, SNS, and JPARK
  fast approaching, the neutron community has the
  potential to undergo a large growth spurt.
• Software is a vital part of neutron scattering,
  and unless the software is both robust and easy
  to use, that growth may be limited.
• Mature packages do exist (McStas, ISAW, DAVE, ),
  and commercial packages are also used (Matlab,
  IDL, Abaqus, IGOR Pro, ) in the analysis
  process. However, groups often use cryptic
  legacy code for at least one step.
• To grow as a community, we need
• a way to cultivate and maintain the valuable
  portions of these legacy codes
• to make legacy and community-standard codes
  interoperable
• define common data structures and interfaces
• stop reproduction of effort
• to allow scientists to concentrate more on
  science by lowering the barriers to software
  engineering

3
There is much to do

• Software is needed to support the massive
  quantity of data that will be produced at modern
  neutron facilities.
• Existing software may be incapable of utilizing
  the full richness of the data that will be
  produced.
• Although the barrier to developing new software
  must be reduced, it is also critical that more
  complex software technologies (i.e.
  high-performance and grid-based computing) are
  enabled.
• Time is short we must use the best existing
  tools to provide a robust solution yet be
  flexible enough to allow for the easy
  substitution of better future solutions.

4
Software User Stereotypes I

• Instrument Scientist
• author of prepackaged and specialized tools
• wants
• portable building and debugging tools
• large toolkit of robust modules and support code
• rapid application development
• GUI builder to compose interactive widgets,
  forms, and wizards
• to focus on supporting the instrument, not
  writing software
• Visiting Scientist
• user of prepackaged and specialized tools
• wants
• UI that is simple to understand easy to use
• reasonable defaults for most choices
• well diagnosed and explained error messages
• intelligently concealed complexity

5
Software User Stereotypes II

• Established Researcher
• coordinator/author/reviewer, designer of new
  applications
• wants
• flexible UI that enables interactive exploration
• access to a comprehensive set of data
  transformations
• access to modeling and simulation packages
• tools to compare outputs of different analyses
• casually useable high-end graphics
• Beginning Student
• user of tools and documentation as learning
  environment
• wants
• well documented interface and modules
• access to a set of standard applications
• flexible UI that enables interactive exploration

6
Software User Stereotypes III

• Analysis Expert
• author of analysis, modeling, or simulation
  software
• wants
• portable building and debugging tools
• large toolkit of robust modules and support code
• easy access to sample data
• to solve physics problems, not software
  engineering problems
• Software Engineer
• binds software to common environment, extends
  software to the framework
• wants
• portable building and debugging tools
• large toolkit of robust modules and support code
• well documented access to the software and
  framework integration layer
• validation, verification, and regression testing
• Framework Maintainer
• maintains and extends the software infrastructure

7
What is DANSE?

• a five-year NSF IMR-MIP software construction
  project
• a collaborative effort between software
  professionals, neutron scattering scientists, and
  facilities
• a software engineering effort
• open-source development environment
• framework for the interoperability of modular
  components
• integration of legacy codes and
  community-standard software
• connectivity to facility databases and software
  repositories
• a scientific endeavor
• to develop software modules for different
  subfields of neutron scattering
• to enhance neutron scattering research and
  facilitate new science
• to build tools for education, collaboration, and
  plausibility assessment
• an integration framework for building data
  analysis, visualization, modeling, and instrument
  simulation tools for all areas of neutron
  scattering

8
The Power of Python

• The fundamental commodity for neutron scattering
  software is found within the cores of time-tested
  community-standard software. Rather than rewrite
  or duplicate this software, we can use python to
  provide an integration path into a common
  language.
• Python is
• a modern object-oriented language
• robust, portable, mature, well-supported,
  well-documented
• easily extendable
• supports rapid application development
• Python scripting enables us to
• compose computations at runtime and discover
  capabilities without recompilation or relinking
• organize large numbers of user-tunable parameters
• Binding Python to other languages (C, Fortran,
  ) allows integration without measurable impact
  on performance or scalability

9
Building a Scientific Toolkit

• Through Python, DANSE will have access to many
  tools
• basic data structures, optimization algorithms,
  numerical libraries
• basic data reduction library obtain I(Q), S(Q),
  S(E), S(Q,E)
• graphical/plotting environments
• IDL, Matlab, Matplotlib, Gnuplot, Grace,
  ParaView, ACIS (AutoCAD),
• instrument simulation
• McStas, VITESS, sample simulation framework,
• materials simulation
• ABINIT, VASP, GAMESS, NWChen, NAMD, CHARMM,
• crystallography
• cctbx, FOX, ObjCryst,
• molecular viewers and format translators
• OpenBabel, Molden, PyMol, ViewMol, DRAWxtl, VMD,
  AtomEye,
• and MORE!
• ISAW, texture analysis (MAUD), SLD calculator,
  scattering intensity,

10
The Power of a Framework

• While a single application can be built
  relatively quickly without using a framework,
  much effort will be spent on error handling,
  logging, UI construction, and other services.
• A software framework provides
• a specification for organization of the software
• a description of the crucial structural elements
  and their interfaces
• a specification of the possible collaborations of
  these elements
• a strategy for the composition of new elements
• flexibility and robustness under evolutionary
  pressures
• services
• life cycle management, logging and monitoring
• network client and server support, authentication
• should not be rewritten for every application,
  but simply reused
• A framework increases reusability decreases the
  development cycle

11
DANSE uses Pyre Framework

• Pyre software architecture
• robust, stable, open-source foundation
• gt75,000 lines of Python 30,000 lines of C
• component-based runtime environment
• components are pre-compiled and connected by the
  user at runtime
• user directs component interconnections using
  visual, script-based, or shell programming
• a set of co-operating abstract services
• framework provides structural girdle
• executive layer manages application life cycle
• applications built from modular components
• components tie software cores to data streams
• UI independent of underlying framework

12
Modularity of Components

• granularity allows reusability of object-oriented
  components
• rebinning application
• modularity provides flexibility and
  extensibility

instrument info
Selector
energy bins
filename
times
filename
Energy
NeXusWriter
NeXusReader
Selector
time interval
Bckgrnd
raw counts
Selector
13
Component Data Flow Paradigm

• scientific analysis codes constitute the cores of
  software components
• components mediate interaction between cores and
  environment
• inherit methods (such as message passing and
  error handling) from environment
• responsible for initialization of programs within
  their component core
• access centralized mechanism for logging status,
  errors, and history
• negotiate data exchanges with XML-based data
  exchange protocols
• components utilize data streams to pass
  information between ports
• interact with executive layer to negotiate
  execution flow
• facilitate physical decoupling of computation
  among distributed resources

14
Component Implementation

• build core engine (Python, Fortran, C, Java,
  Matlab, IDL, )
• legacy or custom code and third-party libraries
• provide life-cycle management and exception
  handling strategy
• construct Python bindings
• select entry points to expose to Python
• modularize entry points to monolithic compiled
  libraries
• cast as a component
• extend and leverage framework services
• describe user-configurable parameters
• provide meta-data that specify the IO port
  characteristics
• test code
• satisfy functional requirements with concurrent
  test development
• utilize interactive runtime testing within Python
  interpreter
• demonstrate integration with other components

15
Building Abstract Applications

• DANSE uses a design pattern that enables the
  assembly of components at runtime under user
  control
• Facilities are named abstract application
  requirements
• Components are concrete named engines that
  satisfy the requirements
• Power of an API
• the application author provides
• a specification of the application facilities as
  part of the application definition
• a component to be used as the default
• the application user can construct scripts that
  create alternative components that comply with
  the facility interface
• the end user can
• configure the properties of the component
• select which component is to be bound to a given
  facility
• Abstraction is required for dynamic and
  distributed applications

16
Visual Programming Interface

• Workflow graphs are a naturally dynamic interface
  due to the correspondence between logical and
  physical descriptions of the computation.
• There are multiple views
  
  of each computation
• data flow
• control flow
• deployment of distributed components
• Should allow interactive editing of component
  state
• access to modify component properties
• dynamic interface generation from
  component-supplied specifications

17
Distributed/Parallel Computing

• Enabled by design
• component framework utilizing data streams
• requirements for building distributed and
  parallel computations nearly the same as those
  for building applications in a visual programming
  interface
• Pyre originally designed to compose and control
  parallel applications
• bindings to mpi
• encapsulation of python interpreter in mpi
• Enable distributed computing with currently
  available technologies
• initial authentication and deployment based on
  ssh scp
• authentication and security using pyre services
• access constrained to user space
• Take advantage of Grid services as they become
  available

18
Broad Scientific Scope

• data reduction and experiment simulation
• diffraction, engineering diffraction, and
  inelastic scattering data reduction
• SANS/USANS and neutron reflectometry data
  reduction
• instrument and microstructure simulation
• modeling
• full profile modeling in real and reciprocal
  space (GSAS, FullProf, PDFFIT)
• finite element modeling (ABAQUS) self-consistent
  modeling
• constrained fitting by use of data from other
  experimental techniques
• 1D/2D model fitting model independent peak
  fitting
• direct modeling of physical systems ab-initio
  modeling
• scattering kernel multiple scattering
• neutron weight correction separation of nuclear
  and spin scattering
• micromagnetic simulations (OOMMF) disordered
  spin dynamics
• chemical spectroscopy dynamics (CLIMAX)

19
Facilitates New Better Science

• better data analysis
• FEM calculations of strains in microstructures
• Monte-Carlo inversions of S(Q,E) to obtain
  parameters of structure and dynamics models
• model refinements with multiple data sets
• integration of theory
• micromechanics using correlations of local
  strains
• phase diagrams from thermodynamic functions
• ab-initio calculations of spin interactions
• soft matter structure using atomic force fields
  guided by diffraction
• experiment planning and execution
• single crystals on chopper spectrometers
• feedback control and real-time assessment
• plausibility testing and contingency planning
• assessment of science/data trends from previous
  data

20
Goals for DANSE

• enable the non-expert, while not hindering the
  expert
• enable distributed and parallel computing as a
  framework service
• create a community-supported open-source neutron
  scattering software framework
• lower the barrier to software development
• provide powerful applications for analysis,
  modeling, and simulation

21
DANSE Project Information

• milestones for the DANSE software
• project start 2006
• beta release 2008
• release 1.0 2009
• transition to the SNS 2010
• documentation, tutorials, and further information
• the DANSE wiki at http//wiki.cacr.caltech.edu/dan
  se
• the Pyre homepage at http//www.cacr.caltech.edu/p
  rojects/pyre
• contacts
• Brent Fultz btf_at_caltech.edu Michael Aivazis
  aivazis_at_caltech.edu
• Simon Billinge, Ersan Üstündag, Paul Butler, Paul
  Kienzle, Ian Anderson
• Michael McKerns mmckerns_at_caltech.edu