A distributed graph processing framework that takes a novel approach by implementing graph algorithms using distributed array and set operations rather than traditional vertex-centric paradigms.
Distributed Grarrph provides high-level abstractions for distributed graph processing by leveraging distributed array and set primitives. Instead of thinking in terms of vertices and edges, algorithms are expressed through common distributed array/set operations.
This approach offers several benefits:
- Cleaner algorithm implementations that focus on the core logic rather than distribution details
- Reusable distributed data structure primitives that can be composed to build complex algorithms
- Natural expression of graph algorithms in terms of bulk operations on collections
- Abstraction from low-level distributed memory management and communication
- Distributed array and set data structures with common operations
- Implementation of standard graph algorithms like BFS and Connected Components
- Integration with KaGen for distributed graph generation
- Benchmarking infrastructure to evaluate scaling performance against other frameworks, i.e HavoqGT and CombBLAS
KaGen generates edge lists with both forward and backward edges, so the actual edge list length is 2m where m is the specified number of edges.