ETD

• distributed-memory
• FastFlow
• hpc
• multi-protocol

To address the ubiquitous increase in data production rates and the storage and computation needs of data-intensive applications, we are recording a shift to distributed computing. New abstractions are needed to enable portability, performance and programmability in the parallel and distributed environment. In order to extend the FastFlow distributed runtime, enabling heterogeneous communications in different parts of the distributed concurrency graph, in this thesis we focus on abstractions targeted at the communication layer, analyzing existing solutions for multi-protocol support. In contrast to existing frameworks, we propose true multi-protocol support, which allows remote nodes to connect using multiple transport protocols (e.g., MPI, TCP, Margo), effectively enabling heterogeneity of compute nodes.