Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

HiPEAC 2020: Energy-aware Task Scheduling in LEGaTO: Low Energy Toolset for Heterogeneous Computing

55 views

Published on

Approach:
-Starting with Made-in-Europe mature software stack, and
optimizing this stack to support energy-efficiency
-Integrated software stack supporting task-based programming model
-Computing on a commercial cutting-edge European-developed
CPU–GPU–FPGA heterogeneous hardware substrate and FPGA-based Dataflow Engines (DFE)
-Three use-cases (Smart home/city, AI, health) to test
the integrated stack

Published in: Science
  • Be the first to comment

  • Be the first to like this

HiPEAC 2020: Energy-aware Task Scheduling in LEGaTO: Low Energy Toolset for Heterogeneous Computing

  1. 1. The LEGaTO project has received funding from the European Union’s Horizon 2020 research and innovation programme under the grant agreement No 780681. www.legato-project.eu Energy-aware Task Scheduling in LEGaTO: Low Energy Toolset for Heterogeneous Computing Miquel Pericàs (miquelp@chalmers.se) Mustafa Abduljabbar (musabdu@chalmers.se) Department of Computer Science and Engineering, Chalmers University of Technology Project Goals One order of magnitude improvement in energy-efficiency for heterogeneous hardware through the use of energy optimized-programming model and runtime. Starting with Made-in-Europe mature software stack, and optimizing this stack to support energy-efficiency Integrated software stack supporting task-based programming model Computing on a commercial cutting-edge European-developed CPU–GPU–FPGA heterogeneous hardware substrate and FPGA- based Dataflow Engines (DFE) Three use-cases (Smart home/city, AI, health) to test the integrated stack Approach 5x decrease in Mean Time to Failure through energy- efficient software- based fault tolerance. Size reduction of the trusted computing base by at least one order of magnitude. 5x times increase in FPGA designer productivity through design of novel features for hardware design using dataflow languages. UseCases Healthcare Will demonstrate not only a decrease in energy consumption but also an increase in healthcare application resilience and security. Machine Learning Will improve energy efficiency by employing accelerators and tuning the accuracy of computations at runtime using CNN and LSTM. IoT for smart homes and cities The LEGaTO project software-hardware framework for the IoT will demonstrate ease of programming and energy savings in smart homes and smart cities applications. Chalmers is extending the XiTAO runtime1 with a low-energy scheduler XiTAO & Energy Nvidia Jetson TX2: 2 Clusters (C0,C1) C0: 2x Denver, 2MB L2 C1: 4x A57 , 2MB L2 FMIN : 345 MHz FMAX : 2035 MHz 1 https://github.com/mpericas/xitao ● Reduces Overheads ● Reduces Parallel Slackness ● Destructive Interference ● Fine-grained parallelism ● Overheads ● Work-time Inflation ● Improves Parallel Slackness ● Bulk creation of parallelism ● Interference-freedom ● Constructive sharing Energy of Energy-Aware Scheduler (EAS) vs Performance Scheduler and Random Work Stealing Scheduler (RWSS) for DAG degrees of parallelism 2, 6 & 10, and Denver/A57 MIN/MAX frequency 0 10 20 30 40 50 60 10 6 2 10 6 2 10 6 2 10 6 2 MAX&MAX MAX&MIN MIN&MAX MIN&MIN EnergyConsumption(mJ) x10000 Energy Comparison of Matrix Multiplication RWSS Perf-based EAS 0 5 10 15 20 25 10 6 2 10 6 2 10 6 2 10 6 2 MAX&MAX MAX&MIN MIN&MAX MIN&MIN EnergyConsumption(mJ) x10000 Energy Comparison of Copy RWSS Perf-based EAS 0 10 20 30 40 50 60 10 6 2 10 6 2 10 6 2 10 6 2 MAX&MAX MAX&MIN MIN&MAX MIN&MIN EnergyConsumption(mJ) x10000 Energy Comparison of Stencil RWSS Perf-based EAS

×