DASH provides generic, efficient multi-threaded data access. It allows data distribution and synchronization in heterogenous environments, implemented by the CoDASH library. The primary use case is the development of task-parallel multi-threaded applications.
DASH fulfills the following requirements:
- Memory efficient
- All threads share the data and copy-on-write
- Single-threaded use does not create any additional copies
- Different data representations may share underlying data, e.g., triangles between a BVH and kd-tree.
- Lock-free and wait-free read access
- Fast write access after first copy-on-write
- Change migration from thread A to B uses simple pointer swaps
- DAG structure
- Access, context mapping and data distribution handles full graphs
- Selective distribution shall be possible (partial data range, partial features)
- See codash
- Data update from node A to B shall only send change delta
- Data discovery shall be transparent
Open an issue if you need help, have a question, want to provide feedback or need advice. We'll try to resolve them in a timely manner.
Building from source
Linux, Mac OS X
git clone https://github.com/Eyescale/Buildyard.git cd Buildyard git clone https://github.com/BlueBrain/config.git config.bluebrain make dash
git clone https://github.com/Eyescale/Buildyard.git cd Buildyard git clone https://github.com/BlueBrain/config.git config.bluebrain [Use CMake UI to configure] [Open Buildyard.sln] [Build 00_Main - dash]