|
| 1 | +============================================================================== |
| 2 | +DistArray 0.4 development release |
| 3 | +============================================================================== |
| 4 | + |
| 5 | +**Documentation:** http://distarray.readthedocs.org |
| 6 | + |
| 7 | +**License:** Three-clause BSD |
| 8 | + |
| 9 | +**Python versions:** 2.7 and 3.3 |
| 10 | + |
| 11 | +**OS support:** \*nix and Mac OS X |
| 12 | + |
| 13 | +What is DistArray? |
| 14 | +------------------ |
| 15 | + |
| 16 | +DistArray aims to bring the strengths of NumPy to data-parallel |
| 17 | +high-performance computing. It provides distributed multi-dimensional |
| 18 | +NumPy-like arrays and distributed ufuncs, distributed IO capabilities, and can |
| 19 | +integrate with external distributed libraries like Trilinos. DistArray works |
| 20 | +with NumPy and builds on top of it in a flexible and natural way. |
| 21 | + |
| 22 | +0.4 Release |
| 23 | +----------- |
| 24 | + |
| 25 | +This is the third development release. |
| 26 | + |
| 27 | +Noteworthy improvements in 0.4 include: |
| 28 | + |
| 29 | +* basic slicing support; |
| 30 | +* significant performance enhancements; |
| 31 | +* reduction methods now support boolean arrays; |
| 32 | +* an IPython notebook that demos basic functionality; and |
| 33 | +* many bug fixes, API improvements, and refactorings. |
| 34 | + |
| 35 | +DistArray is nearly ready for real-world use. The project is evolving rapidly |
| 36 | +and input from the larger scientific-Python community is very valuable and |
| 37 | +helps drive development. |
| 38 | + |
| 39 | +Existing features |
| 40 | +----------------- |
| 41 | + |
| 42 | +DistArray: |
| 43 | + |
| 44 | +* has a client-engine (or master-worker) process design -- data resides on the |
| 45 | + worker processes, and commands are initiated from master; |
| 46 | +* allows full control over what is executed on the worker processes and |
| 47 | + integrates transparently with the master process; |
| 48 | +* allows direct communication between workers, bypassing the master process |
| 49 | + for scalability; |
| 50 | +* integrates with IPython.parallel for interactive creation and exploration of |
| 51 | + distributed data; |
| 52 | +* supports distributed ufuncs (currently without broadcasting); |
| 53 | +* builds on and leverages MPI via MPI4Py in a transparent and user-friendly |
| 54 | + way; |
| 55 | +* supports NumPy-like multidimensional arrays; |
| 56 | +* has basic support for unstructured arrays; |
| 57 | +* supports user-controllable array distributions across workers (block, |
| 58 | + cyclic, block-cyclic, and unstructured) on a per-axis basis; |
| 59 | +* has a straightforward API to control how an array is distributed; |
| 60 | +* has basic plotting support for visualization of array distributions; |
| 61 | +* separates the array’s distribution from the array’s data -- useful for |
| 62 | + slicing, reductions, redistribution, broadcasting, and other operations; |
| 63 | +* implements distributed random arrays; |
| 64 | +* supports ``.npy``-like flat-file IO and hdf5 parallel IO (via ``h5py``); |
| 65 | + leverages MPI-based IO parallelism in an easy-to-use and transparent way; |
| 66 | + and |
| 67 | +* supports the distributed array protocol [protocol]_, which allows |
| 68 | + independently developed parallel libraries to share distributed arrays |
| 69 | + without copying, analogous to the PEP-3118 new buffer protocol. |
| 70 | + |
| 71 | +Planned features and roadmap |
| 72 | +---------------------------- |
| 73 | + |
| 74 | +Near-term features and improvements include: |
| 75 | + |
| 76 | +* MPI-only communication for performance and deployment on clusters and |
| 77 | + supercomputers; |
| 78 | +* array re-distribution capabilities; |
| 79 | +* interoperation with Trilinos [Trilinos]_; |
| 80 | +* expanded tutorials, examples, and other introductory material; and |
| 81 | +* distributed broadcasting support. |
| 82 | + |
| 83 | +The longer-term roadmap includes: |
| 84 | + |
| 85 | +* Lazy evaluation and deferred computation for latency hiding; |
| 86 | +* Integration with other packages [petsc]_ that subscribe to the distributed |
| 87 | + array protocol [protocol]_; |
| 88 | +* Distributed fancy indexing; |
| 89 | +* Out-of-core computations; |
| 90 | +* Support for distributed sorting and other non-trivial distributed |
| 91 | + algorithms; and |
| 92 | +* End-user control over communication and temporary array creation, and other |
| 93 | + performance aspects of distributed computations. |
| 94 | + |
| 95 | +History and funding |
| 96 | +------------------- |
| 97 | + |
| 98 | +Brian Granger started DistArray as a NASA-funded SBIR project in 2008. |
| 99 | +Enthought picked it up as part of a DOE Phase II SBIR [SBIR]_ to provide a |
| 100 | +generally useful distributed array package. It builds on NumPy, MPI, MPI4Py, |
| 101 | +IPython, IPython.parallel, and interfaces with the Trilinos suite of |
| 102 | +distributed HPC solvers (via PyTrilinos [Trilinos]_). |
| 103 | + |
| 104 | +This material is based upon work supported by the Department of Energy under |
| 105 | +Award Number DE-SC0007699. |
| 106 | + |
| 107 | +This report was prepared as an account of work sponsored by an agency of the |
| 108 | +United States Government. Neither the United States Government nor any agency |
| 109 | +thereof, nor any of their employees, makes any warranty, express or implied, |
| 110 | +or assumes any legal liability or responsibility for the accuracy, |
| 111 | +completeness, or usefulness of any information, apparatus, product, or process |
| 112 | +disclosed, or represents that its use would not infringe privately owned |
| 113 | +rights. Reference herein to any specific commercial product, process, or |
| 114 | +service by trade name, trademark, manufacturer, or otherwise does not |
| 115 | +necessarily constitute or imply its endorsement, recommendation, or favoring |
| 116 | +by the United States Government or any agency thereof. The views and opinions |
| 117 | +of authors expressed herein do not necessarily state or reflect those of the |
| 118 | +United States Government or any agency thereof. |
| 119 | + |
| 120 | + |
| 121 | +.. [protocol] http://distributed-array-protocol.readthedocs.org/en/rel-0.10.0/ |
| 122 | +.. [Trilinos] http://trilinos.org/ |
| 123 | +.. [petsc] http://www.mcs.anl.gov/petsc/ |
| 124 | +.. [SBIR] http://www.sbir.gov/sbirsearch/detail/410257 |
| 125 | +
|
| 126 | +.. vim:spell |
0 commit comments