Control for Self-Adaptive, Autonomic Computing Eric RUTTEN Ctrl-A - - PowerPoint PPT Presentation

▶
control for self adaptive autonomic computing
SMART_READER_LITE
LIVE PREVIEW

Control for Self-Adaptive, Autonomic Computing Eric RUTTEN Ctrl-A - - PowerPoint PPT Presentation

Nov 05, 2022 37 likes •198 views

Control for Self-Adaptive, Autonomic Computing Eric RUTTEN Ctrl-A people team @ LIG permanent non-permanent Eric Rutten, CR Inria HdR Neil Ayeb (PhD Orange labs) Gwenal Delaval, MCF UGA Adja Sylla (PhD CEA) Stphane

slide-1
SLIDE 1

Eric RUTTEN

Control for Self-Adaptive, Autonomic Computing

slide-2
SLIDE 2
  • 2

team @ LIG permanent

  • Eric Rutten, CR Inria HdR
  • Gwenaël Delaval, MCF UGA
  • Stéphane Mocanu, MCF INPG

external collaborator

  • Bogdan Robu (Gipsa-lab)

Ctrl-A people

non-permanent

  • Neil Ayeb (PhD Orange labs)
  • Adja Sylla (PhD CEA)
  • Soguy Gueye (post-doc ANR)
  • Chabha Hireche (PhD ANR; Brest)

past non-permanent members

  • Soguy Gueye (post-doc, ANR)
  • Naweiluo Zhou (PhD Labex)
  • Frederico Alvares (post-doc Inria)
  • Julio Cano (post-doc Inria)
  • Mengxuan Zhao (Cifre, PhD)
  • Xin An (PhD, ANR)
slide-3
SLIDE 3

Decision Representation

  • 3

Automated self-adaptation, reconfiguration & regulation reaction to variations load, resources,… large (Cloud, HPC) or embedded (IoT) self-*: deploy, mgmt, healing, protection promising, but challenge in new development method : need for safe automation & separation of concerns Understand and design control for efficiency (e.g; energy) & assurances (e.g.crash avoidance)

Ctrl-A : Control for

Autonomic Computing

Strategy/Policy

Eolas, Grenoble

slide-4
SLIDE 4

Motivation

  • 4
  • Our goal: design languages & model-based methods

validated in target domains

  • Method : attack lack of models & wide range of problems

propose validated generic models

  • Our approach: Software Engineering :
  • Middleware-level instrumentation and architectures,
  • Model-based control (e.g., Discrete Event Systems),
  • Programming support (reactive, components)
  • Targets : HPC, IoT, mid-size grain, heterogeneous

problems : navigation in configurations space

  • Multidisciplinarity : Autonomic Computing, languages

+ control theory, target platforms (HW/MW/SW)

slide-5
SLIDE 5
  • 5

HPC on Dynamically Partially Reconfigurable FPGA

Controlling choices in combinatorial space [ICAC13, ACM TECS16]

Application graph ANR FAmous

par., cond. & seq. branches ANR HPeC

Tasks versions

Size, WCET, power, QoS

Architecture

sleep modes, DVFS, …

Policies power peak,

QoS, surface

Autonomic Computing : Example

ctrlr ctrlr application ctrlr

slide-6
SLIDE 6
  • 6

(Re-)Configurations space (focus : discrete event systems) Interfaces Middleware level API : monitored events, actions Possible behaviors : Automata (parallel, hierarchy) (Hetagon/BZR) Objectives Invariance, reachability, optim. [SefSas18, IEEE TSE16]

Model-based reconfiguration control

Idle Wait e Active r / s r and c

c

/ s

and not c

Idle Active r/s Finish / t

c

e and not c e / t

and c

. . . Idle Wait e r / s r and c

and not c c

/ s A B . . . ci

computing system

ri, ei si, ti

reconfigurable ctrlr actions monitors middleware

slide-7
SLIDE 7
  • 7

excerpt from IEEE ICCAC’17 presentation

  • Design Framework for Reliable Multiple Autonomic

Loops in Smart Environments

  • cooperation with CEA Leti PhD thesis of Adja Sylla
  • transactional middleware Linc
  • applications in Smart office / building
  • methods : Control meets Software Engineering
  • design of safe controller using H/BZR
  • multiple loops to be coordinated

Application to Smart Environments

slide-8
SLIDE 8 |

Generic Autonomic Loop

Abstraction Layer Devices Controller APK M E Legend M Monitoring A Analysis P Planning E Execution K Knowledge data Transactionnal Execution Mechanism data commands commands

Implementation

Transactional Middleware (LINC [Louvel and Pacull, 2014]) Reactive language (Heptagon/BZR [Delaval et al., 2013])

3
slide-9
SLIDE 9 |

Single Loop

Abstraction Layer Devices M E Legend M Monitoring A Analysis P Planning K Knowledge E Execution data commands Transactional Execution Mechanism data commands Controller ADK
  • bjectives

Two kinds of reliability Behavioral Transactional

Other types of controllers

Hand written Based on model checking [Sylla et al., 2015] Based on control theory [Vergara-Gallego et al., 2016]

Single Loop: Limited

6
slide-10
SLIDE 10 |

Coordinated Parallel Loops

Coordinator data data Abstraction Layer data command Devices ADK M E data command data ADK Abstraction Layer data command Devices ADK M E data command Abstraction Layer data command Devices ADK M E data command

Principle

Inhibit an action of a controller Using a coordination variable

Coordinator Design

Manually: using LINC Generation: using Heptagon/BZR

8
slide-11
SLIDE 11 |

Hierarchic Loops

ADK ADK Ctrl4 Ctrl5 Abstraction Layer data command Devices ADK M E data command Abstraction Layer data command Devices ADK M E data command Abstraction Layer data command Devices ADK M E data command Hierarchic Control Loop L1 Ctrl1 Ctrl2 Ctrl3

Motivation Scalability Re-usability Structuring

Abstraction Layer APK M E data data commands commands Abstraction Layer data command Devices ADK M E data command Ctrl1 Abstraction Layer data command Devices ADK M E data command Ctrl2 Ctrl4

Design

Manually: in LINC Generation: in Heptagon/BZR

11
slide-12
SLIDE 12 |

Loop Adaptation

Controllers Abstraction Layer Devices

data commands

ADK M E

data commands

ctrl_A ctrl_B ctrl_C ctrl_D ADK

Principle Controller reconfiguration Conditions related to states

Controller Reconfiguration

In LINC: writing rules In Heptagon/BZR: automata and contract

13
slide-13
SLIDE 13 |

Case Study Description

Office

Sensors: temperature, noise, luminosity, CO2, presence, Agenda Actuators: window, door, lamp, shutter, MV, RAC

Objectives

presence ⇒ luminosity in [500,600] lux and noise < 80 dB presence and temperature > 17 ◦C (> 27 ◦C) ⇒ heat (resp. cool) presence and CO2 > 800 ppm ⇒ ventilation presence and confidential meeting ⇒ office completely closed between two meetings ⇒ quick ventilation not pollution by pollen or outdoor CO2 minimize energy consumption

14
slide-14
SLIDE 14
  • 8
  • Smart home / office Two loops with hierarchical controllers

Lum loop : lamp, shutter TempAirNoise: loop: shutter, window, door, MV, RAC

  • Experimental validation on a model

Lamp Shutter Plugwise circle Raspberry Dongle 070140 EnOcean swicth PTM 210

Case study

slide-15
SLIDE 15
  • 9
  • Goals

tools–supported methods for autonomic controllers design validated by applications in large & small systems

  • Applications

HPC / Cloud infrastructures, FPGA reconfigurable architectures e.g. jLESC joint lab (Inria, Barcelona, ANL, RIKEN @Kobe, …) IoT, smart environments (home, office, building)

  • Perspectives

adaptive control : adaption of the controller itself heterogeneous architectures : e.g. FPGA in data-centers, or comm. networks self-protection : levels of risk/protection, cost w.r.t. functionality

Conclusion

slide-16
SLIDE 16
  • 10

Recent results : Papers Journals

JSS, IEEE TSE, jFACS, ACM TECS, ACM TODAES, FGCS Book chap. SefSas3 (LNCS)

  • Confs. CCTA17; ICAC16,15; ECSA15;

COORDINATION17,14,13; CBSE 14,10 (best paper) Advising 4 PhDs, 3 post-d.

Software Heptagon/BZR, Ctrl-F Projects

4 ANR; 3 Labex ; Orange, CEA; JLESC