Global Task Data-Dependencies in PGAS Applications

摘要: Recent years have seen the emergence of two independent programming models challenging traditional two-tier combination message passing and thread-level work-sharing: partitioned global address space (PGAS) task-based concurrency. In PGAS model, synchronization communication between processes are decoupled, providing significant potential for reducing overhead. At same time, allows to exploit a large degree shared-memory The inherent lack fine-grained in can be addressed through task across process boundaries. this work, we propose use data dependencies describing data-flow synchronize execution tasks created parallel on multiple processes. We present description dependencies, describe necessary interactions distributed scheduler instances required handle them, discuss our implementation context DASH Open image new window framework. evaluate approach using Blocked Cholesky Factorization LULESH proxy app, demonstrating feasibility scalability approach.