https://github.com/nellogan/distributed_compy
Distributed_compy is a distributed computing library that offers multi-threading, heterogeneous (CPU + mult-GPU), and multi-node support
https://github.com/nellogan/distributed_compy
cluster cuda heterogeneous-parallel-programming multi-threading multigpu openmp openmpi
Last synced: 4 months ago
JSON representation
Distributed_compy is a distributed computing library that offers multi-threading, heterogeneous (CPU + mult-GPU), and multi-node support
- Host: GitHub
- URL: https://github.com/nellogan/distributed_compy
- Owner: nellogan
- License: mit
- Created: 2023-11-16T00:59:05.000Z (over 1 year ago)
- Default Branch: master
- Last Pushed: 2023-11-22T20:23:25.000Z (over 1 year ago)
- Last Synced: 2024-08-12T07:10:38.734Z (9 months ago)
- Topics: cluster, cuda, heterogeneous-parallel-programming, multi-threading, multigpu, openmp, openmpi
- Language: Python
- Homepage: https://pypi.org/project/distributed-compy/
- Size: 6.83 MB
- Stars: 2
- Watchers: 1
- Forks: 1
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
Distributed_compy is a distributed computing library that offers multi-threading, heterogeneous (CPU + mult-GPU), and
multi-node (hybrid cluster -- more than one machine with CPU+GPUs) paradigms to leverage the processing power of a
cluster. Cython is used to generate glue code for the core C/C++ functions and provide wrappers to call from Python.
Requires numpy, CUDA toolkit>=2.0, OpenMP, and OpenMPI. Note: this library does not use the popular mpi4py library.
Features:
* Get/set/configure bandwidths of local node or entire cluster whether by supplied numpy array or from binary data files
* Code generator to write temporary binary data files or python files that are to be executed on each node
* Execute mpirun command from master node with default env var or configurable hostfile
* Reduction sum with functionality scaling such as python naive sum, multi-thread reduction sum,
multi-gpu reduction sum, heterogeneous reduction sum, and hybrid heterogeneous reduction sum.
Additional features such as other reduction operations, dot product, matrix multiplication, image processing kernels,
neural networks, and finite element method functions are under consideration for future releases.