SHARQ: Software-Defined Hardware-Managed Queues for Tile-Based Manycore Architectures

摘要

The recent trend towards tile-based manycore architectures has helped to tackle the memory wall by physically distributing memories and processing nodes. Distributed operating systems and applications allow to exploit the increased scalability of such architectures, but still face the data-to-task locality challenge. As inter-tile communication, thread synchronization and data transport often impose significant software overhead on such architectures, many applications would benefit from a more efficient and powerful communication primitive with minimal software involvement. We propose software-defined hardware-managed queues for distributed computing architectures that enable efficient inter-tile communication by leveraging application-specific queues with arbitrarily sized elements. To ensure (remote) processing of queued elements, SHARQ introduces the concept of an optional handler task, which is scheduled by hardware on demand. Queue and memory management, intra- and inter-tile data transfer, and handler task invocation are entirely handled by hardware. Only the dynamic queue creation at runtime is performed in software. As an example use-case, we integrated SHARQ into the MPI library. The evaluation with the MPI-based NAS benchmarks shows a reduction in execution time by up to 48% for the communication intense IS kernel in a 4 × 4 tile design on an FPGA platform with a total of 80 LEON3 cores.