1. Parallel NetCDF Library Development Formerly “Sensor Cloud Integration” Kelsey Weingartner

2. Overview  Background information  Purpose  Project artifacts  Final product  Results  Takeaways

3. NetCDF and MASS NetCDF  Machine-independent format for representing scientific data  Files stores data arranged in variables  Each variable holds an array of data MASS  Library for running a simulation in parallel  Eases the complexity of creating and running 2D and 3D spatial simulations  A simulation is a grid of “Places” that may or may not have “Agents” on them

4. Purpose  Within MASS Make NetCDF file use simple and feasible for MASS Make NetCDF file use simple and feasible for MASS Maintain the benefits of a distributed environment running in parallel. Maintain the benefits of a distributed environment running in parallel.  Real-World Applications Climate change analysis Climate change analysis

5. Artifacts  Summer 2012 Sequential write with NetCDF Sequential write with NetCDF Worst-case parallel performance Worst-case parallel performance  Fall 2013 Best-case parallel performance Best-case parallel performance File creator File creator File creator with parallel write File creator with parallel write File creator with parallel write & read File creator with parallel write & read  Winter 2013 Single instance per processor file creator and parallel writer Single instance per processor file creator and parallel writer Final product Final product

6. Sequential  Each save requires the file to only be opened once  callAll() gathers agent information from each Place  Master node then handles writing to the NetCDF file

7. Parallel - Worst-Case  Each save, the file is opened by every Place object  Master triggers save with callAll()  Place gathers its Agents’ information and writes

8. JavaMPI Parallel Best-Case  Select a NetCDF file to copy  Master node creates a new file with same dimensions  Send an equal portion of data from the chosen file to each node  Each node writes their received array to the newly created NetCDF file

9. Final Product Single Instance per processor file creator and parallel reader/writer  Extends MASS Place  Creates a file for the simulation if none exists  Stores file contents in a buffer to increase read/write speed  Each processor holds the portion of the file relevant to them  A file is only opened by the first writer Place in each partition

10. Results  Sequential write (1 processor): 100x100, 1,000 agents, 1,000 cycles = 225,712.8 msec 100x100, 1,000 agents, 1,000 cycles = 225,712.8 msec  Worst-case parallel write (1 processor): 50x50, 500 agents, 100 cycles 957,590.5 msec 50x50, 500 agents, 100 cycles 957,590.5 msec  MPInetCDF results on a 50x50 file: On 4 processors: 22,114.4 msec / 246,444 bytes = 0.0897 B/msec On 4 processors: 22,114.4 msec / 246,444 bytes = 0.0897 B/msec On 6 processors: 16,470.2 msec / 246,444 bytes = 0.0668 B/msec On 6 processors: 16,470.2 msec / 246,444 bytes = 0.0668 B/msec  RandomWalk using parallel NetCDF (1 processor): 100x100, 1,000 agents, 1,000 cycles = 204,997.2 msec / 472,484 bytes = 0.434 B/msec 100x100, 1,000 agents, 1,000 cycles = 204,997.2 msec / 472,484 bytes = 0.434 B/msec

11. Final Product Results  RandomWalk 100 x 100 grid, 1000 agents, 100 cycles: 7,843.7 msec 100 x 100 grid, 1000 agents, 100 cycles: 7,843.7 msec  RandomWalk with NetCDF 100 x 100 grid, 1000 agents, 100 cycles, writing to file every 20 cycles: 204,997.2 msec 100 x 100 grid, 1000 agents, 100 cycles, writing to file every 20 cycles: 204,997.2 msec Previous settings, but writing to file only once: 69,480.7 Previous settings, but writing to file only once: 69,480.7  Wave2DMASS 100 x 100 grid, 1000 cycles: 16,913.5 msec  Wave2DMASS with NetCDF 100 x 100 grid, 1,000 cycles, writing to file every 50 cycles: 50,422.6 msec Previous settings, but writing to file only once: 22,923.9

12. Future Work On Parallel_NetCDF  D0 array support  Object datatype support  Allow a whole variable to be read/written  Smaller buffer After Parallel_NetCDF   Conference paper for IEEE PacRim Conference

13. Key Lessons  Working with external libraries  Working with limited documentation  Creating and meeting deadlines  Experience with parallel and distributed systems

14. Questions?

15. Intermediate Products File Creators  FileCreator Create uniform 2D or 3D grids Can create NetCDF files with an unlimited dimension.  FileManipulator 1.0 Create uniform 2D or 3D grids Write 1D or 2D arrays of integer  FileManipulator 2.0 Create uniform 2D or 3D grids Read or write whole variable or single value 8 datatypes supported Single Instance Iterations  Single instance per processor reader Create uniform 2D or 3D grids Read or write whole variables 8 datatypes supported

16. Tools Used   Java   Eclipse IDE   JavaMPI   MASS Library   NetCDF Library