Abstract
As many-core systems scale into the hundreds of nodes, design-space exploration becomes an infeasible approach due to the many parameters that need to be optimised to produce a design that fits both the application's requirements and the device constraints. When recent hardware platform problems such as Dark Silicon, device variation and post-manufacture failures are also considered, then a classical design methodology is even harder to achieve. Instead, systems will need to continuously adapt to their operating environment and device parameters. Our previous work has shown that task allocation based on social insect colonies is an effective and efficient approach to tackling the problem of task-to-node mapping in an autonomous and adaptive fashion. In this paper we verify the approach in hardware by implementing the bio-inspired task allocation in a many-core consisting of 100 Microblaze processing nodes connected via a Network on Chip (NoC) with the distributed intelligence embedded within the NoC routers. We show that this adaptive model can be implemented in hardware with a very small hardware overhead of 3% that scales linearly despite the huge number of processing cores on the FPGA chip. Thus this work shows that social-insect inspiration is a effective way of implementing a hardware 'nervous system' that will allow systems to autonomously tackle the problems that ever smaller device implementation technologies bring with them.
Original language | English |
---|---|
Title of host publication | 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016 |
Publisher | IEEE |
ISBN (Electronic) | 9781509042401 |
DOIs | |
Publication status | Published - 9 Feb 2017 |
Event | 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016 - Athens, Greece Duration: 6 Dec 2016 → 9 Dec 2016 |
Conference
Conference | 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016 |
---|---|
Country/Territory | Greece |
City | Athens |
Period | 6/12/16 → 9/12/16 |