Bertenthal

Bertenthal

• 1.
  Social Informatics Data Grid Cyberinfrastructure for Collaborative Research in the Neural, Social and Behavioral Sciences Bennett I. Bertenthal Indiana University [email protected]
• 2.
  Infrastructure for Socialand Behavioral Sciences Goal: Compare, measure and search for patterns in structured, semi- structured, and heterogeneous data sets. Challenge: Integrate information over time, place, and types of data Needs: (1) Data interface (shared datasets & databases) (2) Service interface (shared tools for analysis) (3) Intellectual interface (shared problems & theories)
• 3.
  Primary Objectives • Developprototype of core facility for collecting multiple measures of time-synchronized data • Develop integrated tools for storage, retrieval, annotation, and analyses of multiple data sets at different time scales • Develop scripts for parallelizing code to run on grid clusters
• 4.
  What is SIDGrid?
• 6.
  Social Informatics DataGrid • A general purpose architecture for streaming data applications (e.g., video, audio, time series) • Built on well established database, multimedia and web and grid services standards • Time alignment in distributed heterogeneous datasets – Software and hardware based – Integrated with existing laboratory time stamping and registration techniques • Scalable – Number of datasets – Types of data – Multiple end user applications
• 7.
  Server Client
• 8.
  Client Side
• 9.
  Client Side • Leveragingefforts for annotation and analysis of multimodal data – Familiarity and Interoperability • Elan (Max Planck Institute for Psycholinguistics, The Netherlands) • Talkbank (Carnegie Mellon University, US) • Digital Replay System (Nottingham University, UK) – XML, Java – Cross platform interoperability • Adding SIDGrid functionality to Elan – Minimally intrusive • Avoid complicated co-development w/ELAN team – Browsing SIDGrid data – Additional data types – Upload / Download to SIDGrid server
• 11.
  66 GB 5mov 2 wav … 368 GB 23 mov 6 wav … 5 GB 1 mov 0 wav … 21 GB 3 mov 12 wav … 4 GB 9 mov 1 wav … 4 GB 4 mov 1 wav … 1 GB 0 mov 2 wav … 945 GB 1 mov 66 wav … 8 GB 3 mov 0 wav … 20 GB 13 mov 2 wav …
• 12.
  Server Side
• 13.
  .mov .wav .eaf GB 10 0 0 45 4 30 0 20 2 2 1 3 12 100 9 200 1 1 1 1 6 2 0 12 400 0 1 1001 0 666 1 312 0 0 13 0.1 0 0 0 0.0 18 4 0 66
• 16.
  Search and Query (4,000 projects) • Data Files – Names – Keywords – Attributes (keyword-value) – Date – Type (Elan, Chat) • Contents of Files – Metadata – Tier – Annotations
• 17.
  Server Side • Webservices – Query – Data download / upload • Portal interface – Security – Data and metadata browsing – Preview – Tags, attributes – Projects – Groups – Search – Data transformation using grid resources
• 19.
  Science Gateway
• 21.
  What Is TheTeraGrid? (circa 2006) 75 Teraflops (trillion calculations per second) • 16 Supercomputers - 9 different types, multiple sizes = 12,500 faster than all 6 billion humans on earth each doing one calculation per second • World’s fastest network • Globus Toolkit and other middleware providing single ANL login, application management, data movement, web 30 Gigabits per second to large sites services = 20-30 times major university connections = 30,000 times my home broadband = 1 full length feature film per second LA Starlight Atlanta SDSC TACC NCSA PU IU PSC ORNL
• 24.
  Scripts for RunningJobs on Grid • Matlab (high-level language and interactive environment for peforming computationally intensive tasks) • R (software environment for statistical computing and graphics) • Praat (software for acoustic analysis) • Free Surfer (automated tools for reconstruction of the brain’s cortical surface from structural MRI data) • AFNI (programs for processing, analyzing, and displaying FMRI data) • SUMA (adds cortical surface based functional imaging analysis to the AFNI suite of programs)
• 25.
  Advantages of GridComputing • Vastly expanded computing and storage • Reduced effort as needs scale up • Improved resource utilization; lower costs • Facilities and models for collaboration • Sharing of tools, data, and procedures and protocols • Recording, assessment and reuse of complex tasks
• 27.
  Lessons Learned • Fastprototyping vs production quality software – After one year of development, no product available for user feedback – Optimal design vs practical design • Public vs private website – Need for dissemination – Need for security and protection of user groups and data • Tools for diverse user groups with varying degrees of technical expertise – Non-intuitive interface with minimal user support • Importance of user manuals, technical support, and FAQs • Multiple levels of privacy and confidentiality dictated by type of data and informed consent
• 28.
  If you buildit, will they come? • Dissemination of SIDGrid – Website and movie – Invited workshops at UofC and IU – Pre-conference workshops • Start-up is time consuming – Scale of most projects conducted by social scientists does not justify time to learn web services and tools – Added value for larger, collaborative projects requires shift in goals and organization of research • Resistance to data sharing – Original proposal required that all data stored on SIDGrid servers would be publicly available
• 29.
  Objections to DataSharing • It’s my data! • Protection of confidentiality and anonymity • Need to first establish standards for coding and analysis • Reporting of misleading and confusing findings • Raw data but not coded data should be shared – Annotation and coding is very time consuming and should not become available to others • If availability of web and software tools were contingent on sharing data, most users would opt out
• 30.
  Questions