model-based design and analysis of … · of automotive applications on multicore platforms: ......
TRANSCRIPT
MODEL-BASED DESIGN AND ANALYSIS OF AUTOMOTIVE APPLICATIONS ON MULTICORE PLATFORMS: AN EFFECTIVE APPROACH
Abdoulaye Gamatié LIRMM / CNRS-UM, Montpellier
Journée thématique du GDR SoC-SiP (15 mars 2016) Systèmes Embarqués pour les Transports de Demain
DreamCloud European project (2013 -- 2016)
2 Abdoulaye Gamatié, LIRMM
http://www.dreamcloud-project.org
Resource mapping/allocation heuristics
Examples • Control automation-inspired: feedback loop • Bio-inspired: bee pheromone signalling • Market-oriented heuristics
PE PE
PE PE
PE PE
PE
PE
R
PE
R
R R R
R R R
R
OS
T 1
T 2
T 3
T 4
T 5
T 6
T 7
T 8
T 9
OS
OS
PE PE
PE PE
PE PE
PE
PE
R
PE
R
R R R
R R R
R
OS
OS
OS
T 1
T 2
T 3
T 4
T 5
T 6
T 4
T 5
T 6
T 4
T 5
T 6
T 4
T 5
T 6
Resource allocation layer
T 3
PE PE
PE PE
PE PE
PE
PE
R
PE
R
R R R
R R R
R
OS
OS
OS
T 3
3 Abdoulaye Gamatié, LIRMM
This talk: effective model-based design framework
• Motivation: assessment of mapping heuristics on multicore systems • Real-time properties • Compute and communication performances • Energy consumption
• Proposal: modular and seamless simulation framework • Transaction-level modeling • Cycle-accurate • McSim: Manycore platform Simulation tool-suite
Abdoulaye Gamatié, LIRMM 4
Outline
• Model-based design framework • Sample case-study: automotive application • Improving mapping decision: analysis & prediction • Summary
Abdoulaye Gamatié, LIRMM 5
MODEL-BASED DESIGN Multi abstraction level simulation framework
6 Abdoulaye Gamatié, LIRMM
Automotive technology
7 Abdoulaye Gamatié, LIRMM
Source: http://www.mpsoc-forum.org/previous/2014/slides/jochen_haerdtlein.pdf
Powertrain systems Safety systems Comfort systems
l Multicore µCs (<6) already present in automotive embedded systems
l Higher performance only via Manycore µCs (>6) architectures
Model-based design for automotive applications
8 Abdoulaye Gamatié, LIRMM
• UML-Marte (Modeling and Analysis of Real-Time and Embedded Systems) - http://www.omgmarte.org • general modeling concepts
• AUTOSAR (AUTomotive Open System ARchitecture) -
http://www.autosar.org • single uniform address memory
• Amalthea - http://amalthea-project.org • AUTOSAR-compatible • rich support for multicore systems • support for product-line engineering (variability of modern
motor vehicles)
Application modeling in Amalthea
9 Abdoulaye Gamatié, LIRMM
R2
R1 R3
R4 R0
T0
T1 T3
T2
T6L1
L2
Runnable Label Task Inter-task activation
Instructions in a runnable
o computation: constant delay, delay distribution law
o communication: read (data size) or write (data size)
5ms
10ms
10ms
(i-, i+)
Application modeling in Amalthea (cont’d)
10 Abdoulaye Gamatié, LIRMM
An Eclipse plugin as a user-friendly environment
Application modeling in Amalthea (cont’d)
11 Abdoulaye Gamatié, LIRMM
XML intermediate representation
Multicore architecture modeling
12 Abdoulaye Gamatié, LIRMM
o Core: Amalthea MoC-specific model o Inter-core communication: bus, crossbar, Network-on-Chip (NoC)
Crossbar-based architecture NoC-based architecture
Multicore architecture simulators: a quick survey
13 Abdoulaye Gamatié, LIRMM
Simulator Language Comm. Infrastr. Scalability Accuracy
Real-Time comp. & comm.
gem5 Python/
C++ Various No Cycle No
OVPsim C++ Bus No Functional No
MC-Sim C NoC Yes Cycle No
PREESM C Bus No Instruction No
Simics/GEMS C++ Various No Functional No
Flexus C++ Various Yes Cycle No
McSim SystemC NoC,
Crossbar Yes Transactional, cycle Yes
McSim TLM simulation
14 Abdoulaye Gamatié, LIRMM
** L. Indrusiak and O. dos Santos « Fast and accurate transaction-level model of a wormhole network-on-chip with priority preemptive virtual channel arbitration » In Design, Automation Test in Europe Conference Exhibition (DATE), 2011.
McSim cycle-accurate simulation
15 Abdoulaye Gamatié, LIRMM
** Tran, A. T. and Baas, B. « Noctweak: a highly parameterizable simulator for early exploration of performance and energy of networks on-chip. » Tech. Report ECE-VCL-2012-2, VLSI Comp Lab, ECE Dep. University of California, Davis, 2012.
Parser (internal model generator)
Mapper
Application
NI
Core
NI
Core
NI
Core
NI
Core
NI
Core
Interconnect (NoC)
McSim simulation workflow: example
16
** K. Latif, M. Selva, C. Effiong, R. Ursu, A. Gamatié, G. Sassatelli, L. Zordan, L. Ost, P. Dziurzanski and L. S. Indrusiak. 'Design Space Exploration for Complex Automotive Applications: An Engine Control System Case Study', RAPIDO’16 Workshop.
Abdoulaye Gamatié, LIRMM
17
Typical output results - DemoCar app
Abdoulaye Gamatié, LIRMM
Input app. and platform information
Application parsing output
Simulated timing
information
18
Typical output results - DemoCar app (cont’d)
Abdoulaye Gamatié, LIRMM
Simulated energy
estimation
19
Typical output results - DemoCar app (cont’d)
Abdoulaye Gamatié, LIRMM
Outline
• Model-based design: modular framework • Case-study: an automotive application • Improving mapping decision: analysis & prediction • Summary
Abdoulaye Gamatié, LIRMM 20
CASE STUDY An automotive application
21 Abdoulaye Gamatié, LIRMM
22
A case study: engine control system
Abdoulaye Gamatié, LIRMM
• Amalthea model: 109 tasks, 1239 runnables, 10436 labels
23
A case study: engine control system (cont’d)
Abdoulaye Gamatié, LIRMM
Setup
o McSim TLM NoC simulator
o Two simulated mappings: local-maximized (Loc.) & ZigZag (ZZ)
o Desktop machine: Intel 4-core i5-4670 (3.40GHz)
*APL = Average Packet Latency
24
A case study: engine control system (cont’d)
Abdoulaye Gamatié, LIRMM
01
23
45
67
Deadlines Missed (%) 20002400
28003200
36004000
Execution T
ime (micro
sec)
105
120
135
150
AveragePacketLatency(nanosec)
7x7
7x88x
8
8x99x9
9x10
10x10
7x7
7x8
8x8
8x9
9x9 9x10
10x10
01
23
45
67
Deadlines Missed (%) 20002400
28003200
36004000
Execution T
ime (micro
sec)
50
60
70
80
90
100
110
120
130
140
Energy(mJ) 7x
7
7x8
8x8
8x9
9x99x10
10x10
7x7
7x8
8x8
8x9 9x9
9x10
10x10
Outline
• Model-based design: modular framework • Case-study: an automotive application • Improving mapping decision: analysis & prediction • Summary
Abdoulaye Gamatié, LIRMM 25
IMPROVING DYNAMIC MAPPING DECISIONS Analysis & Prediction
26 Abdoulaye Gamatié, LIRMM
Analysis for behavior prediction
27 Abdoulaye Gamatié, LIRMM
• Behavior prediction modeling
• Why: help dynamic resource allocation algorithms to steer mapping policy
• How: based on historical information, collected from execution traces
• Investigated approaches • Built-in trace analysis (visual outputs of McSim) • Supervised data-mining (not in this talk)
Example of built-in analysis: NoC link load
28 Abdoulaye Gamatié, LIRMM
(ZigZag) (Local-maximized)
Example of built-in analysis: core utilization
29 Abdoulaye Gamatié, LIRMM
(ZigZag) (Local-maximized)
Outline
• Model-based design: modular framework • Case-study: an automotive application • Improving mapping decision: analysis & prediction • Summary
Abdoulaye Gamatié, LIRMM 30
SUMMARY
31 Abdoulaye Gamatié, LIRMM
Overvall orchestration
32 Abdoulaye Gamatié, LIRMM
Mapping heuristics module Mapper
Simulator
Amalthea Parser
Analysis & Prediction
Amalthea application
model
McSim tool-suite: https://github.com/DreamCloud-Project
Mc
Sim
MODEL-BASED DESIGN AND ANALYSIS OF AUTOMOTIVE APPLICATIONS ON MULTICORE PLATFORMS: AN EFFECTIVE APPROACH
Abdoulaye Gamatié
*** Joint work with: C. Effiong, K. Latif, L. Ost, G. Sassatelli, M. Selva and R. Ursu, L. Zordan, P. Dziurzanski, L. S. Indrusiak...