SMC2 Planning

SMC2 planning investigated an integrated system for memory-centric computing. The NSF-supported planning effort (opens in a new tab) proposed combining near-memory data processors with an extensible remote memory pool to reduce the impact of memory accesses in graph-mining applications.

The design brings together architecture, software and hardware interfaces, programming models and compilers, and performance models. Its research directions include intelligent prefetching, speculative data movement, support for near-memory function calls, and communication between near-memory processors.

Related graph-processing work examines how data organization and architecture affect memory access. Learning-Based Reordering (opens in a new tab) adaptively identifies and reorders frequently accessed graph data. Skyway (opens in a new tab) combines a direct path for graph state data with memory-side buffers tailored to different graph data types.