Overview of Swallow --- A Scalable 480-core System for Investigating the Performance and Energy Efficiency of Many-core Applications and Operating Systems

TL;DR

Swallow is a scalable 480-core many-core system designed for studying performance and energy efficiency of applications and operating systems, featuring low-latency interconnects and a distributed memory architecture.

Contribution

This paper introduces Swallow, a scalable open-source many-core system with a novel architecture enabling efficient experimentation with applications and OS.

Findings

Provides 240 GIPS performance at low power per instruction

Demonstrates scalable software development with a distributed OS (nOS)

Shows practical use cases like neuron modeling and shared memory overlay

Abstract

We present Swallow, a scalable many-core architecture, with a current configuration of 480 x 32-bit processors. Swallow is an open-source architecture, designed from the ground up to deliver scalable increases in usable computational power to allow experimentation with many-core applications and the operating systems that support them. Scalability is enabled by the creation of a tile-able system with a low-latency interconnect, featuring an attractive communication-to-computation ratio and the use of a distributed memory configuration. We analyse the energy and computational and communication performances of Swallow. The system provides 240GIPS with each core consuming 71--193mW, dependent on workload. Power consumption per instruction is lower than almost all systems of comparable scale. We also show how the use of a distributed operating system (nOS) allows the easy creation of scalable software to exploit Swallow's potential. Finally, we show two use case studies: modelling neurons and the overlay of shared memory on a distributed memory system.