Gerhard P. Fettweis – Coordinator W. Lehner, W. Nagel, C. Baier, D. Plettemeier 29.05.2015 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing 3 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing The Research Idea of HAEC: Highly Adaptive Energy-Efficient Computing
Motivation 3 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing 200% Energy Consumption with Reference to 2010 Grid 150% Non-ICT Current trend: 10x in 20 years Current trend: 10x in 20 years 100 % Non-ICT >50% of ICT ? 50 % 90% 2030 grid ICT 10% HAEC Challenge: 0 % enabling a massive-magnitude reduction! 2010 2030 Energy Proportionality/Efficiency 4 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Center for Information Services and High Performance Computing (ZIH) Measurement at June 20, 2008 Goal: Minimizing Energy Percentage by Multi-Layer SW/HW Rel. Energy Consumption Adaptivity Rel. Computing Load Time of Day
The Communication Bottleneck of Parallel Computing 5 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing P = 16 P arallelism = 2 C = 120 C omm. Links = 1 P = 32 P = 4 C = 496 C = 6 P = 64 P = 8 C = 2016 C = 28 Highly Adaptive Energy-Efficient Computing High-Rate Inter-Chip Communications 6 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Optical Interconnect � adaptive analog/digital circuits for e/o transceiver � embedded polymer waveguide � packaging technologies (e.g. 3D stacking of Si/III-V hybrids) � 90 ° coupling of laser R di I Radio Interconnect � on-interposer/on-package � antenna arrays � analog/digital beam steering and interference minimization � 100Gb/s � 25 GHz channel @ 200GHz carrier � 3D routing & flow management
Energy Optimization 7 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Today’s energy optimizations Are considered on a component level � Mostly in isolation � Unaware of applications and contexts � Goal of project group B Address energy-efficiency at all system levels in an integrated holistic manner � Ranging from the OS to software components to virtual machines � HAEC B – Closed Energy-Control Loop 8 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Energy-Optimization Control Loop � Measurement Software Architecture � Energy-adaptive SW stack � Analysis & verification � Strategic regulation Monitoring Control � Adaptation Run-time Optimization � Focus on energy efficiency Energy-Adaptive Software Architecture � Using context information � Global QoS optimization with “Energy-Utility Functions” Hardware Architecture � Cross-layer adaptation: � New communication channels application - system - hardware
Example of Adaptivity by Software 9 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing � Three in-memory database services � All are using a scan to access their data � Under-utilized cores and interconnects Example of Adaptivity by Software 10 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing How to optimize? 1) Exchange software implementations (dbscan � index scan) 2) Migrate database service 3) Migrate memory 4) Turn off unnecessary interconnects Hardware and software adaptations Necessary for optimal energy-efficiency/proportionality
Example of Adaptivity 11 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing How to gain (more/less) utility from (more/less) energy ? 1) Partition memory and activate additional sockets and interconnects 2) Add more computing power and activate necessary interconnects Trade Energy for Utility/Capacity Highly Adaptive Energy-Efficient Computing 12 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Application Simulation Middleware Flexible Software & Information Processing Runtime Prototyping Operating System Energy Control Processor Loop Devices & Flexible Hardware Circuits HAEC Box Interconnects
15 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Highlights of Phase I (2011-2015) HAEC A 20 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Wireless ADC Butler LNA Base- System Integration Mixer Matrix PA DAC band HAEC Architecture (Packaging) HAEC Software Device Modelling Optical LD TIA Base- MZM LDD band PD
Selected HAEC A Highlights 21 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Embedded TIA with waveguides Transimpedance Adaptivity: Amplifier (TIA) Bandwidth vs 90 Gbit/s DC power 0,23 pJ/bit System Integration � 50% Energy 0,16 mm 2 HAEC Architecture (Packaging) HAEC Software Device Modelling Optical LD TIA Base- MZM LDD band PD Selected HAEC A Highlights 22 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Wireless ADC Butler LNA Base- System Integration Mixer Matrix PA DAC band HAEC Architecture (Packaging) HAEC Software Device Modelling A05 Bow-tie antenna (1mm x 0.6 mm) A01 200 GHz 18 mW A01 200 GHz A05 A01 24GSps 3bit LNA (low noise amplifier) 18 mW mixer Half clover- single core leave antenna flash ADC (1.2mm x 0.7 mm)
Selected HAEC A Highlights: Artemis Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing slide 23 slide # should Shunt-based not contain instrumentation for “slide” energy measurements Energy modeling in a state-of-the-art join subgroup subselect server leftfetchjoin projectdelta leftfetchjoinPath 1,2 relative energy consumption 1,15 1,1 1,05 1 1300 1800 2300 2800 3300 processor frequency (MHz) Energy monitoring Highlights Software HAECubie slide 24 Selected HAEC B Highlights: HAECubie Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Resource Management (B04) sensor data Knowledge Plane system information Database (B05) System Management update HAECubie trade-offs data aggregated data Service Execution (B06) Energy-Utility Analysis (B03) context synopses es coordination Software Architecture (B01) adaptations Context Recognition (B02) (B02) context
The Outlook: The Box in 2030+ 30 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing Assume 64K processors per chip double-sided 4x4 chip-stacks on board 160x chips stacked in 3D 4x boards in a box Box in 10x10x10 cm 3 (1 liter) � 10 8 processors! 1.6K processor + 16K memory chips � 10 4 x performance of today! HAEC Box Embedded Everywhere 31 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing
High-Level Highlights / Conclusions 32 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing � Adaptivity � Energy proportionality � Energy efficiency � Revolutionary approach to computing hardware/software � Impact onto more than computing applications � Chance for scalability of the box and the software approach � Chance for transfer 33 Collaborative Research Center 912: HAEC � Highly Adaptive Energy-Efficient Computing
Recommend
More recommend