D ISTRIBUTED S YSTEMS - The Next Grand Challenge in Embedded System - PowerPoint PPT Presentation

D ISTRIBUTED S YSTEMS - The Next Grand Challenge in Embedded System Design Jan M. Rabaey Donald O. Pederson Distinguished Prof. Director FCRP MultiScale Systems Center (MuSyC) Scientific Co-Director Berkeley Wireless Research Center University of California at Berkeley I NTEL , F EBRUARY 23 2011

The Swarm and the Cloud TRILLIONS OF CONNECTED DEVICES Infrastructural THE core CLOUD THE SWARM [J. Rabaey, ASPDAC’08]

The New Moore’s Law Still … Improve functionality per unit cost to create whole new application areas, But in a brand new setting Cloud Wireless & Computing Personal Devices Internet Mainframes PCs Immersive User Experiences Ubiquitous Sensing 1970 1990 2000 1980 2010-

The Swarm Perspective Moore’s Law Revisited: Scaling is in number of connected devices, no longer in number of transistors/chip The functionality is in the swarm! Resources can be dynamically provided based on availability It’s A Connected World Time to Abandon the “Component” -Oriented Vision [MuSyC 2009]

One Vision: CyberPhysical Systems Linking the Cyber and Physical Words [H. Gill, NSF 2008]

Another One: BioCyber (?) Systems Linking the Cyber and Biological Worlds Examples: Brain-machine interfaces and body-area networks

The Cloud and the Swarm Distributed Sense and Control Challenges Failure to Address in Fundamental and Cohesive Way will System Slow Down or Prohibit Metrics Adoption (ENERGY) Modeling/ Security/Tru Abstractions st Complexity Verification Robustness/ Reliability Run-time Management / Diagnostics

It’s All About Energy Energy among most compelling Smart grid concern of distributed IT platform and its applications. Mobiles Avionics Human-centric systems OUR VISION: Distributed Sense and Control Systems to Dynamically Enforce Energy-Proportionality

Business as Usual Will Not Do The mantra’s of two decades of low -power design: slow, simple, many, dedicated, adaptive While some opportunities are left, concepts now commonly exploited The end of voltage and energy scaling !? 1 1 12x 0.1 Energy (norm.) 0.1 Total Switching Leakage 0.01 0.01 0.3V 0.001 0.001 0 1 0.2 0.4 0.6 0.8 1.2 V DD (V) In Need of Novel Unless novel devices are adapted soon … Architectural Ideas

The Golden Opportunity Energy-Proportional Computing Actual Throughput Ideal Energy-efficiency of most systems decreases under reduced loads Power Courtesy: L. Barroso, Google

Computation and Energy Power� Actual� Ideal� How we design systems DOING NOTHING (or LITTLE) WELL How nature designs systems Throughput� Energy efficiency of most systems degrades under reduced load conditions [ * Term coined by L. Barroso, Google]

A Generic Concept  Conceive and Enable Systems that are Energy-Proportional over Large Throughput Power Range.  Applies to all aspects of the IT Platform! Actua l Ideal DOING NOTHING (or LITTLE) WELL Throughput Not the case in today’s systems (computing, storage, communication)

The Big Picture Attention-Optimized Computing/Communication Maximization “Matching computation to desired utility” Utility Hugely Scalable Platforms “Providing computation/computation at the optimal energy” A Closed Loop System

The Cloud/Swarm Challenge  Trade off computation and communication  in light of limited energy, communication and, computational resources  so that desired utility is reached  under highly variable conditions and loads Requires scalable distributed optimization strategy

The “Playground” Distributed Resources Sensing Communication Storage Energy Computation Actuation (Spectrum) The Swarm/Cloud Operating System - Dynamically trading off resources A continuously changing The Swarm/Cloud Services and Applications alignment (environment, density, activity) Utility Maximization “What matters in the end is the utility delivered to the user”

A DDRESSING T HE C HALLENGES

Focus Center Research Program Features  Multi-university teams  Focus on topics where evolutionary R&D is insufficient  Emphasis on discovery; long-range time horizon  Large-scale effort (~ $7M per center annually)  Equal cost sharing between industry & government  Access to relevantly trained graduate students “The (SRC) focus center program is designed to create a nationwide, multi-university network of research centers that will keep the United States and U.S. semiconductor firms at the front of the global microelectronics revolution.” Craig R. Barrett Retired Chairman of the Board, Intel Former Chair, Semiconductor Technology Council Recent Chair, FCRP Governing Council 17

M U S Y C IN A N UTSHELL 20 Faculty Distributed over 10 US Universities Create comprehensive and systematic solution the Grand Goal: distributed multi-scale system design challenge. “Energy - smart” distributed systems, that  Are deeply aware of balance between energy availability and Grand Challenge: demand  Adjust behavior through dynamic and adaptive optimization at all scales of design hierarchy. Common Core: SCS Theme Distributed sense and control systems . Target: Airborne Platforms (Avionics) Exploring the multi-scale space: LSS Theme SSS Theme Large- scale “energy - intensive” Small- scale “energy - frugal” systems systems Target: Human-centered networks for Target: Data centers augmented sensing (e.g. BMI)

T HE M U S Y C T EAM SCS LSS SSS Including experts in petascale computing, networking, control, signal processing, information theory, avionics and neuro-engineering

THEME 1: DISTRIBUTED SENSE AND CONTROL SYSTEM METHODOLOGY (A. SANGIOVANNI- VINCENTELLI) Address challenges in complex distributed control systems by employing structured and formal design methodologies that seamlessly and coherently combine various dimensions of multi-scale design space, and that provide appropriate abstractions to manage inherent complexity. Case study: Avionics Complexity

SCS D RIVERS A ND M ETRICS Large Airborne Platforms Today Power sources/sinks Electric distribution Control system • # power sources ~1 • # loads ~100 In Line with DARPA META Program • peak power ~ 400kW Tomorrow Reduction of development time of complex, distributed control systems by 2X through increased use of formal methods for specification, design and verification. Reduction of the number of faults that require the system to be taken out of service for inspection or repair by 2 X, through the increase used of onboard models and • # power sources ~ 10 dynamic reconfiguration to provide enhanced • # loads ~1000 fault tolerance. • peak power ~ 4MW

SCS HIGHLIGHT: FORMULATED DESIGN FLOW FOR DISTRIBUTED AVIONICS SYSTEMS Current State of the Art Contributors: E. Lee, R. Murray and ASV Incremental conservative design Our Approach • Steady state worst case power draw • 2x overdesign results in weight penalty (STRONG impact on META I and II BAA) Platform-based design enables architecture Power System exploration (tradeoff weight, stability, …) Architecture Dynamics, control, Control System communication latency Dynamics problems Architecture addressed in all layers Redesign identified in verification Ptolemy, Metro tools Hardware, Software, Communications latency enable robust design Communications impacts stability of complex dynamical systems Robust design for distributed control system Realistic Test Benches under development Collaboration with UTC (HS), IBM and Raytheon

THEME 2: LARGE-SCALE SYSTEMS (T. SIMUNIC- ROSING) Realize distributed closed-loop power-management strategies that result in “energy - intensive” large-scale systems to be orders of magnitude more energy-efficient , while ensuring that mission-critical goals are met. To be accomplished by employing holistic multi-scale solution including all components of the system at multiple hierarchy levels. Target: Data centers “Doing nothing well”

LSS DRIVERS AND METRICS METRIC: Datacenter Energy Efficiency SOLUTION : Distributed and hierarchical management that ensures that energy is only consumed if, when and where needed . Enable “ energy- proportional” computing, and to “do nothing well ” in Datacenters and Cloud Computing Barroso & Hölzle, 2009

LSS H IGHLIGHT : E NERGY - AWARE LOAD SCHEDULING Contributors: Katz, Snavely, Rosing, NSF GreenLight Cooling-aware management Energy Supply Information Building/Facility Manager Tasks Energy Aware Workload Model/ Cluster Workload B SLAs Predictor Manager Scheduler Energy Consumption Application Resource Footprint

THEME 3: SMALL-SCALE SYSTEMS (D. JONES) Explore absolute bounds of energy-efficiency and miniaturization in “energy - frugal” human-centric distributed IT systems , through distributed management strategy that dynamically and adaptively selects correct operational point corresponding to varying application needs in terms of accuracy or resolution. Target: Augmented sensing in humans (BMI)

SSS DRIVERS AND METRICS KEY METRIC: UTILITY/ENERGY Utility Maximization • Define system performance in terms of user/application relevant utility • Dynamically optimize algorithms and platforms to maximize utility Explore, analyze, and implement advanced closed-loop learning systems in brain-machine interfaces In collaboration with UCB Neuroscience and UCSF Neurosurgery