Difference between revisions of "Summer student projects"
(→finished =) |
(move finished projects into proper section) |
||
| Line 20: | Line 20: | ||
* Cactus: support running tests in parallel [Perl -- proposed as student project at NCSA, Roland] | * Cactus: support running tests in parallel [Perl -- proposed as student project at NCSA, Roland] | ||
| − | == | + | ==Finished projects== |
| + | * <strike>Carpet: write a single output files per MPI rank rank rather than one per group or one per variable</strike> [C++ -- Roland] | ||
| + | * <strike>GetComponents: track component inter-dependencies to handle components being checked out into directories provided by others [Perl -- Roland]</strike> | ||
* Cactus/MoL: investigate performance of Adams-Bashforth time integrators in MoL [none -- Roland, Ian, Erik; summer student at NCSA] | * Cactus/MoL: investigate performance of Adams-Bashforth time integrators in MoL [none -- Roland, Ian, Erik; summer student at NCSA] | ||
* Cactus/MoL: add startup code for Adams-Bashforth time integrators to MoL [C -- Roland, Erik, Ian; summer student at NCSA] | * Cactus/MoL: add startup code for Adams-Bashforth time integrators to MoL [C -- Roland, Erik, Ian; summer student at NCSA] | ||
| Line 26: | Line 28: | ||
** RH: it is probably not sufficient, but is necessary to get things working. I am still not sure if the buffer zones cannot be removed during checkpoint recovery. It may also be possible without. | ** RH: it is probably not sufficient, but is necessary to get things working. I am still not sure if the buffer zones cannot be removed during checkpoint recovery. It may also be possible without. | ||
** this project is done, unfortunately no speedup. | ** this project is done, unfortunately no speedup. | ||
| − | |||
| − | |||
| − | |||
| − | |||
Revision as of 12:45, 24 October 2017
Projects for Summer Students
This page lists some projects for summer students, suitable e.g. for an REU student or Google Summer of Code students. Most of these projects are both self-contained and open-ended, and could be extended at will.
(Entry format: Subproject: Title [Required knowledge -- potential supervisor])
Computer Science
- Simfactory: Automate archiving of simulation results [Python -- worked on as summer project at NCSA Roland, Erik]
- Simfactory: add and test code to resubmit jobs automatically based on Cactus status return [Python -- Roland]
- ICH: already implemented in SimFactory 3
- Carpet: Collect performance statistics of parallel HPC simulations, output and display these [C++, MPI -- Erik]
- Carpet: Design and implement a tree data structure for (integer) bounding boxes [C++ -- Erik]
- McLachlan: Improve performance of OpenCL code on CPUs [OpenCL, assembler -- Erik]
- McLachlan: Improve performance of OpenCL code on GPUs [OpenCL, GPUs -- Erik]
- Cactus: Implement an interface between Cactus and PETSc [C/C++, parallel elliptic solver -- Ian, Erik, Eloisa, Roland]
- Carpet: populate Carpet::point_classification [C++ -- Roland, Erik; partially done by Roland]
- Cactus: support HDF5 data in tests [Perl, python -- worked on as as student project at NCSA, Roland]
- Cactus: support running tests in parallel [Perl -- proposed as student project at NCSA, Roland]
Finished projects
Carpet: write a single output files per MPI rank rank rather than one per group or one per variable[C++ -- Roland]GetComponents: track component inter-dependencies to handle components being checked out into directories provided by others [Perl -- Roland]- Cactus/MoL: investigate performance of Adams-Bashforth time integrators in MoL [none -- Roland, Ian, Erik; summer student at NCSA]
- Cactus/MoL: add startup code for Adams-Bashforth time integrators to MoL [C -- Roland, Erik, Ian; summer student at NCSA]
- ICH: is this useful without corresponding code in Carpet to remove the buffer zones?
- RH: it is probably not sufficient, but is necessary to get things working. I am still not sure if the buffer zones cannot be removed during checkpoint recovery. It may also be possible without.
- this project is done, unfortunately no speedup.
