OPEN PROBLEMS IN MULTI- MODAL SCHEDULING THEORY FOR THERMAL-RESILIENT MULTICORE SYSTEMS Nathan Fisher, Masud Ahmed, and Pradeep Hettiarachchi Department of Computer Science Wayne State University CoPaRTS 1
Open Problems in Multi-Modal Scheduling theory for thermal-Resilient multicore Systems
Jan 11, 2016
Compositional & Parallel Real Time Systems. CoPaRTS. Open Problems in Multi-Modal Scheduling theory for thermal-Resilient multicore Systems. Department of Computer Science Wayne State University. Nathan Fisher , Masud Ahmed, and Pradeep Hettiarachchi. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
OPEN PROBLEMS IN MULTI-MODAL SCHEDULING THEORY FOR THERMAL-RESILIENT MULTICORE SYSTEMS
Nathan Fisher, Masud Ahmed, and Pradeep Hettiarachchi
Department of Computer ScienceWayne State University
CoPaRTS1
Thermal Resiliency [Motivation]:
2
Heart Regulate
Status Transmit
Data Log
Extensive Exercises
When surrounding temperature increases: Reduce CPU thermal dissipation for the device
safety. Drop non-essential tasks on demand.
Designing hard-real-time systems with predictable degradation in a dynamic thermal
environment.
Required: Processor/Task Control Framework
Control Framework [Multi-Modal Overview]
3
System Hardware SpecificationSystem Software Specification
M1M3M4M2M5
Temperature/Power and Workload ModelsControl System Design
(Task 1)(Task 2)
(Task 5)
(Task 4)
(Task 3)
Priority vary over time
High Priority
Less Priority
Critical
Less Priority
Need formal models for mode changes in software and hardware.
Control Framework [Modes]Real-time performance modes: M(1),…,M(q)
Each M(i) is a collection of sporadic tasks {τ(i)
j}j=1…n and a periodic resource
(i)=((i),(i)).
Possible to model processor with two power
levels
Pact: active power
Pinc: inactive power
Π(i)
Θ(i) Θ(i)Θ(i)
timeHow do we deal with changes of operating modes?
Control Framework [Mode-Change Requests]
5
M(i) M(j) M(k)
)( ij jk)()( jjN mcrk
Mode-Change Request
Transition
time
time
mcrk-1
)( i
Mode:
…
)( j
)( j
)( jTasks:
)( i )( j )( k)( kResource:
……
Assumption: Mode-change request occurs at period boundaries
Control Framework [Task Mode-Change Semantics]6
tk-1 +
tk
M(j) M(k)
)(ij
)( jk
Immediately Aborted Tasks α(ij)
Non-Aborted Tasks
Unchanged Tasks τ(ij).
X
X
Control Framework [Multi-Modal Schedulability Analysis]7
Mj)(ijkt Mi
Busy Interval “BI5”
Busy Interval “BI1”
Busy Interval “BI2”
Busy Interval “BI3”
Busy Interval “BI4”
kt
Intra-Mode Schedulability Conditions
Previous Work: Multi-modal Uniprocessor Schedulability Analysis for Periodic Resources
[ESTIMedia 2011, ACM-TECS 2014]
Inter-Mode Schedulability Conditions
Control Framework [Schedulability Analysis]
8
Mjijkt Mi kt
Previous Work: Multi-modal Uniprocessor Schedulability Analysis for Periodic Resources
[ESTIMedia 2011, ACM-TECS 2014]
Open Problem: Multi-modal Multiprocessor Global Schedulability Analysis for a
Compositional Resource Model
Multiprocessor Compositional Resource Models9
Potential Models: Multiprocessor Periodic Resource (MPR)
Model: each resource characterized (i)=((i),(i), m(i)) [Shin et al, ECRTS ‘08]
Parallel Supply Function (PSF) Model [Bini et al, RTSS ‘09]
Maximum Concurrency
Π(i)
time
P1
P2
.
.
.
Pm -1(i)
Pm +1(i)
Pm (i)
.
.
. . . .Θ(i)
Challenges/Issues
Transient backlog over multiple mode changes.
“Carry In/Out” calculations. Time Complexity. Relation to mixed-criticality scheduling? “Optimal” resource parameters:
What is a good definition for thermal resiliency?
How do you calculate efficiently?
10
Related Documents
