Extending the Performance and Energy-Efficiency of Shared Memory Multicores with Nanophotonic Technology

Abstract

As the number of cores increases exponentially on a single chip, the design and integration of both the on-chip network facilitating intercore communication, and the cache coherence protocol for enabling shared memory programming have become critical for improved energy-efficiency and overall chip performance. With traditional metal interconnects facing stringent energy constraints, researchers are currently pursuing disruptive solutions such as nanophotonics for improved energy-efficiency. Cache coherence in multicores can be enforced effectively by snoopy protocols; however, broadcasting every cache miss can limit the scalability while consuming excess energy. In this paper, we propose PULSE, a nanophotonic broadcast tree-based network for snoopy cache coherent multicores. To limit the energy-penalty from broadcasting (and thereby splitting) optical signals, we direct the optical signal from the external laser such that only the subset of requesters can receive the optical signal. Furthermore, as cache blocks are shared by a few cores, we propose a multicast version of PULSE called multi-PULSE that predicts the sharers' for each L2 miss and morphing the broadcast to a multicast network. We evaluate the energy and performance using CACTI and SIMICS on 16-core and 64-core versions of PULSE and multi-PULSE for Splash-2, PARSEC, and SPEC CPU2006 benchmarks and compare to electrical networks, optical networks, and another cache filtering techniques. Our results indicate that PULSE outperforms competitive electrical/optical networks by 60 percent in terms of execution time, and multi-PULSE reduces average energy from 10 to 80 percent even with a few mispredictions.

Scalability of Broadcast Performance in Wireless Network-on-Chip

“…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].…”

“…Laying down the transmission line and the required transceivers takes a significant portion of area, and less power than the wireless option given that, in transmission lines, signals are guided instead of radiated and losses are reduced. On the other hand, a all-optical 64-core NoC based on global broadcast buses is proposed in [26]. The estimated power is high compared to the rest of alternatives, but provides a huge broadcast bandwidth of 320 bits per clock cycle.…”

“…For instance, multicast methods have been extended to 3D topologies enabled by vertical stacking [24]. 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].…”

Using Frequency Scaling on Virtualized Memory in Cloud Datacenters

“…Its key ideaology is that adaptability in energy conservation can be achieved through the integration of a dynamic disk scheduling scheme and power management in parallel disk systems. In [19], the authors proposed a broadcast tree-based network for snoopy cache coherent multicores in memory shared systems. Those studies were effective for reducing the energy consumption of storage systems at hardware level.…”

Optimal Power Efficient Photonic SWMR Buses

“…There have been some recent proposals [24], [10], [15] that propose to dynamically allocate bandwidth to groups of nodes based on their declared or predicted usage. For example, Zhou et al [24] use a tree based power waveguide to distribute optical power to the nodes.…”