Broadcast-Enabled Massive Multicore Architectures: A Wireless RF Approach

Abstract

Abstract-Broadcast has been traditionally regarded as a prohibitive communication transaction in multiprocessor environments. Nowadays, such constraint largely drives the design of architectures and algorithms all-pervasive in diverse computing domains, directly and indirectly leading to diminishing performance returns as we reach the manycore era. Novel interconnect technologies could allow to revert this trend by offering, among others, improved broadcast support even in large-scale chip multiprocessors. In this position paper, the prospects of wireless on-chip communication technologies pointing towards low-latency (a few cycles) and energy-eff cient (a few pJ/bit) broadcast are outlined. This work also discusses the challenges and potential impact of adopting these technologies as key enablers of unconventional hardware architectures and algorithmic approaches, in the pathway of signif cantly improving the performance, energy eff ciency, scalability and programmability of manycore chips.

Scalability of Broadcast Performance in Wireless Network-on-Chip

“…Synchronization has been particularly affected by this fact, as it has become expensive by default and can degrade performance by a 40% in average and by seven times in extreme cases like streamcluster albeit representing a small fraction of the code [28]. The main reason is that these functions would greatly benefit from global communication schemes [7], [29], implying that in spite of the recent coherence-aware proposals [12], [30], multicast support will become a key design point for NoCs. Also, and in spite of the recent NoC-aware works in architecture [31], [32], this implies that avoiding multicast may become unaffordable at some point.…”

“…However, globally shared-medium schemes will only be considered as serious contenders provided that their scalability is demonstrated. Among the different ways to implement such strategy [23], [25], [26], we chose to focus on the wireless RF paradigm due to its inherent broadcast capabilities and potential scalability [7].…”

“…A major concern is that, despite its limited support, multicast and broadcast may play a decisive role in future architectures [7]. Cache coherency, arguably the main source of on-chip communication in shared memory multiprocessors, is currently implemented through directorybased schemes that use multicast to invalidate cache blocks on a shared write.…”

“…These range from directory-like schemes like the AMD HyperTransport [10], to classical snooping protocols adapted to unordered networks [11], [12], and token coherence [13]. In message passing, collective primitives such as MPI_Allgather or MPI_Allreduce use multicast and are widely employed in parallel algorithms [7]. Finally, novel computing paradigms could be also multicast-driven: multiprocessors emulating neural network architectures communicate their cores through multicast "spike" messages [14].…”

“…Further, the service of multicast traffic through dedicated broadcast channels by means of global RF transmission lines and nanophotonic waveguide bundles has been inspected [25], [26]. Finally, on-chip wireless communication stands as one of the most promising options given its inherent broadcast capabilities [7], [23], [27].…”