Optimal All-to-All Personalized Communication in All-Port Tori

All-to-all personalized communication, also known as complete exchange, is one of the most dense communication patterns in parallel computing. In this paper, we propose new indirect algorithms for complete exchange on all-port ring and torus. The new algorithms fully utilize all communication links and transmit messages along shortest paths to completely achieve the theoretical lower bounds on message transmission, which have not be achieved among other existing indirect algorithms. For 2D r × c ( r % c ) all-port torus, the algorithm has time complexities of optimal transmission cost and O(c) message startup cost. In addition, the proposed algorithms accommodate non-power-of-two tori where the number of nodes in each dimension needs not be power-of-two or square. Finally, the algorithms are conceptually simple and symmetrical for every message and every node so that they can be easily implemented and achieve the optimum in practice.

A Generic and Extensible Spidergon NoC

The Globally Asynchronous Locally Synchronous Network on Chip (GALS NoC) is the most efficient solution that provides low latency transfers and power efficient System on Chip (SoC) interconnect. This study presents a GALS and generic NoC architecture based on a configurable router. This router integrates a sophisticated dynamic arbiter, the wormhole routing technique and can be configured in a manner that allows it to be used in many possible NoC topologies such as Mesh 2-D, Tree and Polygon architectures. This makes it possible to improve the quality of service (QoS) required by the proposed NoC. A comparative performances study of the proposed NoC architecture, Tore architecture and of the most used Mesh 2D architecture is performed. This study shows that Spidergon architecture is characterised by the lower latency and the later saturation. It is also shown that no matter what the number of used links is raised; the Links×Diameter product permitted by the Spidergon architecture remains always the lower. The only limitation of this architecture comes from it-s over cost in term of silicon area.