Building and Operating Large-Scale SpiNNaker Machines

Access to files

•   FULL-TEXT.PDF (pdf)

Abstract

SpiNNaker is an unconventional supercomputer architecture designed to simulate up to one billion biologically realistic neurons in real-time. To achieve this goal, SpiNNaker employs a novel network architecture which poses a number of practical problems in scaling up from desktop prototypes to machine room filling installations.SpiNNaker's hexagonal torus network topology has received mostly theoretical treatment in the literature. This thesis tackles some of the challenges encountered when building `real-world' systems. Firstly, a scheme is devised for physically laying out hexagonal torus topologies in machine rooms which avoids long cables; this is demonstrated on a half-million core SpiNNaker prototype. Secondly, to improve the performance of existing routing algorithms, a more efficient process is proposed for finding (logically) short paths through hexagonal torus topologies. This is complemented by a formula which provides routing algorithms with greater flexibility when finding paths, potentially resulting in a more balanced network utilisation.The scale of SpiNNaker's network and the models intended for it also present their own challenges. Placement and routing algorithms are developed which assign processes to nodes and generate paths through SpiNNaker's network. These algorithms minimise congestion and tolerate network faults. The proposed placement algorithm is inspired by techniques used in chip design and is shown to enable larger applications to run on SpiNNaker than the previous state-of-the-art. Likewise the routing algorithm developed is able to tolerate network faults, inevitably present in large-scale systems, with little performance overhead.

Layman's Abstract

SpiNNaker is a supercomputer designed to simulate neural networks, such as those that make up the brain. Unlike biological experiments, simulations running on SpiNNaker allow neuroscientists to control and observe the behaviour of these networks while avoiding the need for animal experiments. Though it is extremely unlikely that SpiNNaker will ever `think', it may lead to a better understanding of the functions and disorders of the brain.Like most supercomputers, SpiNNaker is made up of many smaller computer processors interconnected by a network. When the largest planned machine is completed, SpiNNaker will contain over one million computer processors -- each responsible for simulating several hundred neurons -- together capable of simulating neural networks similar in scale to a cat's brain.This thesis makes three contributions towards the construction and operation of full-sized SpiNNaker machines. Firstly, I devised a new way of organising the physical components which make up a large SpiNNaker machine so that only short cables are required to construct its network making SpiNNaker cheaper and easier to build. Secondly, I developed a method that makes it possible for the processors in SpiNNaker to communicate reliably, even when some connections are faulty -- an unavoidable situation in practice. Finally, I adapted a technique normally used to design computer chips to assign neural models to SpiNNaker's processors. This method ensures connected neurons are assigned to nearby processors inside SpiNNaker's network meaning that signals between neurons have to travel less distance and are not as likely to get in each other's way. This makes it possible to simulate larger and more complex neural networks.

Keyword(s)

Computer Science; Fault tolerance; Graphs; Hexagonal Torus Topology; Networks; Place and Route; Simulated annealing; Software; SpiNNaker; Spiking neural networks; Supercomputer; Topology

Bibliographic metadata

Institutional metadata

Record metadata