Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
1HPSC IRT Overview
HPSC | Nebulae Workshop, Keck Institute For Space Studies, Caltech | 08.26.19
Reinventing the Role of Computing in SpaceRichard J. Doyle, NASA Project Manager, High Performance Spaceflight Computing (HPSC)Jet Propulsion Laboratory, California Institute of Technology
© 2019 California Institute of Technology. Government sponsorship acknowledged.
Nebulae – In-Space Computing
Space-based computing has not kept up with the needs of current and future missionsNASA has some unique requirements
• Deep space, long duration, robotic and human missions • Higher performance, smaller spacecraft and lower cost
‒ Onboard science data processing‒ Autonomous operations
• Extreme needs for low power and energy management, efficiency, fault tolerance and resilience
Based on alignment of interests and plans, USAF and NASA entered into a programmatic partnership for a joint investment in flight computing
2
Strategic Importance of Flight Computing Future Mission Needs
Pre-Decisional Information – For Planning and Discussion Purposes Only 2
GCD FY18 Mid Year ReviewNebulae – In-Space Computing 3
Com
puta
tiona
l Rat
e (MI
PS)
Intel Motorola 680X0 PowerPC MissionsLaunch Year
68020/3368030/50
68040/40
68060/75
80386/33
80486/25
80486/50
Pentium/60
Pentium Pro/150 200Pentium II/233
Pentium II/450Pentium III/450
PPC601/80PPC601/110
PPC604/132PPC603e/133
150200
266333
300 400 PPC7400/450
Galileo CDS(1802) Mars Observer EDF
(1750A)
Clementine HKP(1750)
Mars Global Surveyor(1750A)
Mars Pathfinder Rover(80C85)
Cassini(1750A)
Mars Pathfinder AIM(RAD6000)
Deep Space 1(RAD6000)
Stardust(RAD6000)
1
10SIRTF
(RAD6000)
Deep Impact(RadLite750)
PPC7455/1000PPC7441/700
Pentium 4/2530
Pentium 4/2000 PPC7470/1250
86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
MSL RAD750 III
Rad-hard components are always at least 2 generations behind commercial State Of The Art
MultiCore
Multi-Core Regime
COTS Single-Core Era
Flight (Rad-hard) Single-Core
Pentium 4/3400
RAD5545
Dual Core LEON FT
Intel Corei7/5820X
Intel Corei7/4790k
Intel Corei5/4670k
LEON 4CORE FT
OPERA Maestro
HPSC
History of Spaceflight Avionics and Processors
Credit: R. SomePre-Decisional Information – For Planning and Discussion Purposes Only
4Nebulae – In-Space Computing
Human Spaceflight (HEOMD) Use Cases1. Cloud Services2. Advanced Vehicle Health Management3. Crew Knowledge Augmentation Systems4. Improved Displays and Controls5. Augmented Reality for Recognition and
Cataloging6. Tele-Presence7. Autonomous and Telerobotic
Construction8. Automated Guidance, Navigation, and
Control (GNC)9. Human Movement Assist
Science Mission (SMD) Use Cases1. Extreme Terrain Landing2. Proximity Operations / Formation Flying3. Fast Traverse4. New Surface Mobility Methods5. Imaging Spectrometers6. Radar7. Low Latency Products for Disaster
Response8. Space Weather9. Science Event Detection and Response10. Immersive Environments for Science
Operations / OutreachHigh value and mission critical applications identified by NASA scientists and engineers
HPSCNASA Use Cases
Pre-Decisional Information – For Planning and Discussion Purposes Only
Onboard Data Product GenerationDust Devils on Mars
Dust devils are scientific phenomena of a transient nature that occur on Mars– They occur year-round, with seasonally variable frequency– They are challenging to reliably capture in images due to their dynamic nature– Scientists accepted for decades that such phenomena could not be studied in real-time
New onboard Mars rover capability (as of 2006)• Collect images more frequently, analyze onboard to detect events, and only
downlink images containing events of interestBenefit
• < 100% accuracy can dramatically increase science event data returned to Earth• First notification includes a complete data product
Spirit Sol 543 (July 13, 2005)
5Credit: T. Estlin. B. Bornestein, A. Castano, J. Biesiadecki, L. Neakrase, P. Whelley, R. Greeley, M. Lemmon, R. Castano, S. Chien and MER project team
Nebulae – In-Space Computing
Computation Category
Mission Need Objective of Computation
Flight Architecture Attribute
Processor Type and Requirements
Vision-based Algorithms with Real-Time Requirements
• Terrain Relative Navigation (TRN)• Hazard Avoidance• Entry, Descent & Landing (EDL) • Pinpoint Landing
• Conduct safe proximity operations around primitive bodies• Land safely and accurately• Achieve robust results within available timeframe as input to control decisions
• Severe fault tolerance and real-time requirements• Fail-operational• High peak power needs
• Hard real time / mission critical• Continuous digital signal processing (DSP) + sequential control processing (fault protection)• High I/O rate• Irregular memory use• General-purpose (GP) processor (10’s – 100’s GFLOPS) + high I/O rate, augmented by co-processor(s)
Model-Based Reasoning Techniques for Autonomy
• Mission planning, scheduling & resource management • Fault management in uncertain environments
• Contingency planning to mitigate execution failures• Detect, diagnose and recover from faults
• High computational complexity• Graceful degradation• Memory usage (data movement) impacts energy management
• Soft real time / critical• Heuristic search, data base operations, Bayesian inference• Extreme intensive & irregular memory use (multi-GB/s)• > 1GOPS GP processor arrays with low latency interconnect
High Rate Instrument Data Processing
High resolution sensors, e.g., SAR, Hyper-spectral
• Downlink images and products rather than raw data • Opportunistic science
• Distributed, dedicated processors at sensors• Less stringent fault tolerance
• Soft real time• DSP/Vector processing with 10-100’s GOPS (high data flow)• GP array (10-100’s GFLOPS) required for feature ID / triage
HPSCNASA Flight Computing Drivers
6
Nebulae – In-Space Computing7
Eigen-App Throughput DSP GP P LP MC1 1-10 GOPS X X X X2 1-10 GOPS X X X X3 10-50 GOPS X X X X X4 10-50 GOPS X X X X5 10-50 GOPS X X X X6 10-50 GOPS X X X7 50-100 GOPS X X X X X8 50-100 GOPS X X X X9 50-100 GOPS X X X X
10 50-100 GOPS X X X
• Bundled requirements that represent groups of key-cross cutting applications
• Derived by selecting low power applications from full applications set and grouping by throughput, processing type, mission criticality
App to Eigen-App Mapping DSP GP P Mission Critical LP
Throughput = 1-10 GOPSAutonomous Mission Planning X X X XDisaster Response X X XHyspiri X X X
Throughput = 10-50 GOPSFast Traverse X X X X XExtreme Terrain Landing X X X X XAdept X XOptimum Observation X X X XSpace Weather X X XRobotic Servicing X X X XCloud Service X X XAdvanced ISHM X X XAutonomous and Telerobotic Construction X X X X
Througput = 50-100s GOPSHyperspectral Imaging X X X XRADAR Science X X XRADAR EDL X X X XAutomated GN&C X X X XHuman Movement Assist X X X XCrew Knowledge Augmentation X XImproved Displays and Controls X X X XAugmented Reality X X XTelepresence X X X
HPSCDerived Eigen-Apps
KEY• DSP – Digital Signal Processing• GP – General Purpose Processing• P – Parallelizable• Mission Critical – Requires Additional Fault Tolerance• LP – Max Power Available for Processor Chip <6W
Credit: R. Some7
GCD FY18 Mid Year ReviewNebulae – In-Space Computing
HPSC – Technology OverviewReinventing the Role of Computing in Space
8
• HPSC offers a new flight computing architecture to meet the needs of NASA missions through 2030 and beyond. Providing on the order of 100X the computational capacity of current flight processors for the same amount of power, the multicore architecture of the HPSC chiplet provides unprecedented flexibility in a flight computing system by enabling the operating point to be set dynamically, trading among needs for computational performance, energy management and fault tolerance. This kind of operational flexibility has never been available before in a flight computing system.
• HPSC has been conceived to be highly extensible. Multiple chiplets can be cascaded together for more capable computing, or HPSC can be configured with specialized co-processors to meet the needs of specific payloads and missions.
• HPSC is a technology multiplier, amplifying existing spacecraft capabilities and enabling new ones. The HPSC team anticipates that the chiplet will be used by virtually every future space mission, all benefiting from more capable flight computing.
Nebulae – In-Space Computing
0 5 10 15 20
5
250 BAE RAD750
BAE RAD5545
10152025303540 Maestro
Boeing (HPSC Study)Honeywell (HPSC Study)
BAE (HPSC Study)
HPSC 8-Core Chiplet
HPSC 14-Core Chiplet
Worse
Better
Power (W)
Perfo
rman
ce (G
OPS
)
HPSC 8-Core Chiplet - scaled
HPSC 8-Core Chiplet - extended
HPSC Target
9
It’s time to move beyond 1990’s technology and reinvent the role of computing in space systems
HPSCPower vs. Performance
9
GCD FY18 Mid Year ReviewNebulae – In-Space Computing
HPSCPerformance
Key Performance ParametersPerformance
Parameter State of the Art Threshold Value Project Goal Estimated Current Value
Computational 350 MOPS 9 GOPS 15 GOPS 15 GP GOPS +Up to 100 SIMD GOPS
Power Scalability 10W 10W scalable to <1W 7W scalable to <1W 9.1W scalable to TBD W
Performance / Power 20 MOPS/W 0.9 GOPS/W 1.5 GOPS/W 1.5 GP GOPS/W + TBD SIMD
GOPS/W
I/O Bandwidth 800 Mbps 40 Gbps (4 SRIO) 60 Gbps (6 SRIO) 240Gps (6 SRIO @ up to 40Gbs/port)
Fault ToleranceCustomized redundancy hardware
Application / middleware SW-level methods with timing, coverage and power advantages
Hardware / System SW-level methods with improved efficiency over application / middleware methods
Application / middleware SW-level methods with timing, coverage and power advantages
Extensibility Generally not available Cascade chiplets via SRIO Cascade chiplets via SRIO
transparently Cascade chiplets via SRIO
Note: Working estimates from PDR; high resolution updates by CDR.
10
GCD FY18 Mid Year ReviewNebulae – In-Space Computing
HPSCPartners
NASA Team
USAF Team
Boeing Team
Sponsor
NASA GSFCProject Management, Procurement, System Engineering, Middleware, SMD Mission Applications
NASA JSCHEOMD Mission ApplicationsSPLICE Project
JPLProject Management, System Engineering, Middleware, SMD Mission Applications
NASA STMD (HQ)NASA Game Changing Program OfficeChiplet Baseline and NASA SE/Oversight Sponsor
USAF/SMCUSAF/DoD Mission ProgramsChiplet QualificationChiplet Options Sponsor
Boeing SSEDChiplet Development Contractor
University of MichiganARM Design Sub-Contractor
USC Information Sciences InstituteSSW Development Sub-Contractor
TAPOTrusted Applications Program Office
COSMIACUSAF/DoD Qualification Subcontractor
Boeing BDSChiplet Development Contractor
AFRL/RVUSAF Space Electronics Program OfficeMiddleware Sponsor
11
Nebulae – In-Space Computing
HPSCFlight Demonstration
NASA Commercial Lunar Payload Systems (CLPS) Lunar Lander Technology Demonstration
Pre-Decisional Information – For Planning and Discussion Purposes Only 12
GCD FY18 Mid Year ReviewNebulae – In-Space Computing
HPSCMission Infusion
• The use cases provide a basis for requirements and an infusion framework for board-level solutions tailored to different mission application categories.
Human Spaceflight
Surface Systems High Data Rate Instruments
Landing Systems
Deep Space
CubeSats, SmallSatsPre-Decisional Information – For Planning and Discussion Purposes Only 13
GCD FY18 Mid Year ReviewNebulae – In-Space Computing 14
FUTURE OPPORTUNITY With the advent of more capable and flexible flight computing (High Performance Spaceflight Computing – HPSC) and onboard analytics (machine learning, data science), the timing is right to explore and develop scaled and newly enabled scenarios for science / engineering autonomy
STAKEHOLDERS• Mars orbiter and surface mission applications integrating autonomy, flight computing
and deep space communications / networking
• Additional NASA and non-NASA mission applications
Task PI
1 COSMIC – Content-based On-board Summarization to Monitor Infrequent Change
Lukas Mandrake
2 MAARS – Machine Learning-based Analytics for Automated Rover Systems
Hiro Ono
3 MOSAIC – Mars On-Site Shared Analytics, Information, and Computing
Joshua Vander Hook
MilestonesAlgorithm development for onboard science event detection, terrain classification, etc.
Architecture development for sharing computing and memory resources across primary / secondary Mars assets
Benchmarking of algorithms and analytics against emulated HPSC + Emu capability
NEMO-referenced and Fetch Rover-referenced demonstrations of science and engineering autonomy
Integrated technology demonstration of autonomy, computing, and networking in a Mars environment
Lukas Mandrake, Hiro Ono, Joshua Vander Hook, Rafi Some, Rich Doyle
0 5 10 15 20
5
250 BAERAD750
10
15
20
25
30
35
40 Maestro
HPSC8-CoreChiplet
Power(W)
Perfo
rmance(G
OPS)
HPSC8-CoreChiplet - scaled
HPSC8-CoreChiplet- extendedHPSC
Target
OBJECTIVES
Prepare for a more capable future flight computing environment by validating enhanced and additional scenarios for science and engineering autonomy
1) Optimize the science return of Mars orbiters via onboard analytics such as change detection
2) Enable continuous science and more efficient and reliable operations for Mars surface assets
3) Benchmark autonomy software / analytics / against HPSC and co-processor (e.g., Emu) capability
4) Validate the feasibility and utility of sharing computing and memory resources in a planetary environment
BedrockSand
Sand
Bedrock
Hill
Noon 2pm
4pm
MinimumdistancepathGoalStart
10amNoon
2 pm
10am
10amNoon
2am
Onboard Analytics for Mars Exploration Enabled by HPSC
Credit: L. Mandrake, H. Ono, J. Vander HookPre-Decisional Information – For Planning and Discussion Purposes Only
Nebulae – In-Space Computing
Coordinated Science ObservationsNetwork Orbiter Assisted Operations
Network Support for Lunar Operations
Vision
DTN enables new mission services and operational concepts
Applicable to deep space, near-Earth, and lunar missions
DTN Concepts for Space-based Networking by
Credit: J. WyattPre-Decisional Information – For Planning and Discussion Purposes Only 15
Nebulae – In-Space Computing
Observatories of the FutureAutonomy Priority Example (8X)
16
Earth-ObservingConstellations
Integrated autonomy architecture • Event detection and response • Formation flying• Onboard data product generation
Earth-Observing Observatories of the Future
16Credit: J. Hyon
Nebulae – In-Space Computing
Enabling Space Exploration and Scientific Discovery by Emplacing Scalable Infrastructure
• Traditionally, space missions have been approached as having self-contained resource envelopes to accomplish specific objectives
• The concept of space platforms coming and going, supported by emplaced and accumulating infrastructure, already applies for Earth orbit and more and more, to Mars exploration
• This KISS study is about looking beyond, farther into the solar system, and exploring the possibilities for resources in the area of computing, data storage, and networking
17
What are the possibilities?