real-time operating systems laboratoryredwood.snu.ac.kr/wordpress/wp-content/uploads/2018/08/... ·...
TRANSCRIPT
Web page: http://redwood.snu.ac.kr, Tel: +82-2-880-8370
Real-Time Operating Systems Laboratory
서울대학교 전기정보공학부 교수 (1995.9~현재)
서울대학교 미래융합기술최고위과정 주임교수 (2017.3~현재)
서울대학교 자동화시스템 공동연구소 연구소장 (2016.9~현재)
한국공학한림원 정회원 (2018.1~현재)
서울대학교 융합과학기술대학원 부원장 (2012.10~2014.10)
서울대학교 융합과학기술대학원 융학과학부 학부장 (2012.10~2014.10)
가헌과학기술재단 석좌교수 (2008.4~2017.3)
한국자동차공학회 사업이사 (2010.1~2011.12, 2013.1~2015.12, 2017.1~2017.12)
전기전자시스템/ITS 부문위원회 부회장 (2012.1~현재)
IEEE RTCSA Steering Committee 위원장 (2013. 8 ~2016.8)
교수님 주요 경력 주요 연구 업적
Linux Kernel Scheduling for Multicore Systems
Fair-share scheduling on asymmetric multicore systems
big Cluster LITTLE Cluster
Linux Kernel
SVR-CFS
Inter-cluster Task Migrator
… …
Per-core
Scheduler
Per-core
Scheduler
𝑹𝝆Estimator
𝑹𝝆Estimator
…
Per-core
Scheduler
Per-core
Scheduler
𝑹𝝆Estimator
𝑹𝝆Estimator
…
big big LITTLE LITTLE
Intra-cluster SVR Balancer
(big Cluster)
Intra-cluster SVR Balancer
(LITTLE Cluster)
Task Information Measured Data
• Linaro’s scheduling framework의 inter-cluster task migrator 사용• 제한된 power budget에서성능을높이기위해,
- compute-intensive task들을 big cluster에서수행- I/O-intensive task들을 LITTLE cluster에서수행
Inter-cluster Task Migrator
• Fairness 달성을위해,
- 현재까지의 SVR이낮은 task에게더많은 CPU 자원할당- 현재까지의 SVR이높은 task에게더적은 CPU 자원할당
• SVR(scaled virtual runtime)은 asymmetric multicore system에서한 task의진행률을나타내기위한지표
Intra-cluster SVR Balancer
• 𝑅𝜌 Estimator를통해현재각코어의 relative performance 계산
• 계산된결과를기반으로각 task의 SVR 계산• Fairness 달성을위해,
- 각코어에서수행중인 task들의 SVR 값이동일하도록스케줄링
Per-core Scheduler
Reducing memory interference latency of safety-critical applications via memory request throttling and Linux cgroup
CPU
Memory Controller
OtherDMA
Devices
Memory
GPU
SystemBus
Critical ApplicationsNormal Applications
자원 경쟁 발생
Excessive Memory Contention in Mixed-Criticality System Solution Architecture
Linux
Kernel
Number ofOutstanding
Requests
Excessive Memory Contention
Predictor
HardwareCPUMemory GPU
cgroup
cgroup
Core
cgroup Controller
Memory Request RateController
OtherControllers
CPUFreq
Governor
CPUScheduler
Memory ControllerDriver
GPU DeviceDriver
CPUFreq
Architecture Driver
CurrentlyRunningTasks
Memory RequestRate Throttling
Requests
Occurrence of ExcessiveMemory Contention
Task List ofCritical and
Normal cgroup
(1) the start ofa critical task chain
write()
System Call
The Start Task ofCritical Task Chain
Parent and Child TaskSender and Receiver Task
write()
System Call
The End Task ofCritical Task Chain
…… …
(2) the extension ofa critical task chain
via an IPC call
(3) the extension of a critical task chain via
creating a task
(4) the end ofa critical task chain
Classification Requests
IPCSystem Call
fork()
System Call
Critical Task ChainManager
Programming Language and Runtime forFuture Autonomous Vehicles
Splash: stream processing language for real-time and embedded AI
Hardware
Real-Time Pub-Sub Communication Middleware(based on DDS Specification)
Linux Kernel
CPU Memory Storage GPU
Automotive Ethernet CAN
Sensor Data Control Data
Real-Time Stream Processing Framework
Splash Applications
Scene Recognition Path Planning Vehicle Control
… … …
Splash RuntimeSplash Program Example: 3D Object Detection
Kernel Techniques for Active Resource Management
In-Depth Kernel Optimization Techniques
Improving interactivity via cross-layer resource control and scheduling
for Linux/Android smartphones
Android Application Framework
Linux Kernel
…
Frame Buffer/dev/input/event*
Dalvik VM Instance
Main
Task Surface
Flinger
Task
Input
Dispatcher
Task
Input
Reader
TaskDalvik VM
Instance
Main
Task Binder
Task
User Touch Input Identifiers of the user-interactive tasks
Binder IPCConsolidated
Image
Input Data
Worker
Task
Foreground App
…
Other Apps
Promote priorities of the worker tasks in a user-interactive task chain Restore priorities of the promoted tasks
…User-Interactive Task Chain
Event Hub
Message QueueInput Channel Buffer
Layer
Device
MemoryOutput Device
(LCD)
Input Device
(Touch Screen) CPU
Solution Architecture
Lag#!&%
Problem Android may demonstrate a
perceptible lag in a GUI when an interactive task is running with multiple CPU-bound applications
Kernel Hackers possessing Solid Theoretical Basis
문제 선정, 정보 수집, 구체화, 해결책 제안, 검증에 이르는 전 과정을자기 주도적으로 수행할 수 있는 “자기완결적 문제해결능력” 보유자
탄탄한 이론적 배경 지식을 바탕으로 실용적인 시스템 소프트웨어 기술을체득하여, 이를 실제 산업에 적용할 수 있는 능력 보유자
※ RTOS Lab의졸업생들은학계를비롯한삼성전자, Google, Amazon, Oracle, Pure Storage, BlackRock 등글로벌기업에진출하여활약중임