schedulability analysis of synchronous digraph real-time tasks · introduction real-timetaskmodels:...
TRANSCRIPT
Schedulability Analysis ofSynchronous Digraph Real-Time Tasks
Morteza Mohaqeqi, Jakaria Abdullah, Nan Guan, Wang Yi
Uppsala University
ECRTS 2016
Introduction
Real-Time Task Models:
Liu & Layland
sporadicmultiframe (MF)
generalized MF (GMF)
non-cyclicGMF
recurring branching (RB)
Digraph (DRT)
Proposed by M. Stiggeet al. (2011)Real-time tasks withdifferent job types
v5 v6
v7
30
9
10
8
Synchronous Digraph Real-Time Tasks - 1/18 - Mohaqeqi, Abdullah, Guan and Yi
Introduction
Real-Time Task Models:
Liu & Layland
sporadicmultiframe (MF)
generalized MF (GMF)
non-cyclicGMF
recurring branching (RB)
Digraph (DRT)
Proposed by M. Stiggeet al. (2011)Real-time tasks withdifferent job types
v5 v6
v7
30
9
10
8
Synchronous Digraph Real-Time Tasks - 1/18 - Mohaqeqi, Abdullah, Guan and Yi
The Digraph Real-Time (DRT) Task Model
Job Types• WCET• Relative deadline
Conditional flow(Branch)
v1
v2
v3
v4〈4, 15〉
〈1, 10〉
〈2, 5〉
〈1, 20〉
minimuminter-release〈WCET , deadline〉
15
40
20
2510
t0 5 10 15 20 25 30 35 40 45 50 55 60
v1 v2 v4 v3
t0 5 10 15 20 25 30 35 40 45 50 55 60
v1 v2 v1
Synchronous Digraph Real-Time Tasks - 2/18 - Mohaqeqi, Abdullah, Guan and Yi
The Digraph Real-Time (DRT) Task Model
Job Types• WCET• Relative deadline
Conditional flow(Branch)
v1
v2
v3
v4〈4, 15〉
〈1, 10〉
〈2, 5〉
〈1, 20〉
minimuminter-release〈WCET , deadline〉
15
40
20
2510
t0 5 10 15 20 25 30 35 40 45 50 55 60
v1 v2 v4 v3
t0 5 10 15 20 25 30 35 40 45 50 55 60
v1 v2 v1
Synchronous Digraph Real-Time Tasks - 2/18 - Mohaqeqi, Abdullah, Guan and Yi
The Digraph Real-Time (DRT) Task Model
Job Types• WCET• Relative deadline
Conditional flow(Branch)
v1
v2
v3
v4〈4, 15〉
〈1, 10〉
〈2, 5〉
〈1, 20〉
minimuminter-release〈WCET , deadline〉
15
40
20
2510
t0 5 10 15 20 25 30 35 40 45 50 55 60
v1 v2 v4 v3
t0 5 10 15 20 25 30 35 40 45 50 55 60
v1 v2 v1
Synchronous Digraph Real-Time Tasks - 2/18 - Mohaqeqi, Abdullah, Guan and Yi
Outline
1 A Review on DRT
2 Synchronous DRT
3 Schedulability Analysis
4 Conclusion
Synchronous Digraph Real-Time Tasks - 3/18 - Mohaqeqi, Abdullah, Guan and Yi
Synchronous DRT
Synchronized Release
Synchronous Digraph Real-Time Tasks - 4/18 - Mohaqeqi, Abdullah, Guan and Yi
Semantics
v1
v2
v3
v4
4
1
1
2
s1
15
40
25
2510
Task T1:
v5 v6
v7
2 1
1
s1
20
9
10
8
Task T2:
t0 5 10 15 20 25 30 35 40
t0 5 10 15 20 25 30 35 40T1
T2
v1 v2 v3
v5 v6blocked
Synchronous Digraph Real-Time Tasks - 5/18 - Mohaqeqi, Abdullah, Guan and Yi
Semantics
v1
v2
v3
v4
4
1
1
2
s1
15
40
25
2510
Task T1:
v5 v6
v7
2 1
1
s1
20
9
10
8
Task T2:
t0 5 10 15 20 25 30 35 40
t0 5 10 15 20 25 30 35 40T1
T2
v1 v2 v3
v5 v6blocked
Synchronous Digraph Real-Time Tasks - 5/18 - Mohaqeqi, Abdullah, Guan and Yi
Overview
AssumptionsUniprocessorPreemptive schedulingFixed priority
Contributions� Schedulability analysis� Heuristics for better efficiency
Synchronous Digraph Real-Time Tasks - 6/18 - Mohaqeqi, Abdullah, Guan and Yi
Overview
AssumptionsUniprocessorPreemptive schedulingFixed priority
Contributions� Schedulability analysis� Heuristics for better efficiency
Synchronous Digraph Real-Time Tasks - 6/18 - Mohaqeqi, Abdullah, Guan and Yi
Outline
1 A Review on DRT
2 Synchronous DRT
3 Schedulability Analysis
4 Conclusion
Synchronous Digraph Real-Time Tasks - 7/18 - Mohaqeqi, Abdullah, Guan and Yi
DRT Schedulability
v1 v2
v3
〈1〉 〈2〉
〈1〉
15
9
10
8
t0 5 10 15 20 25 30 35
v1 v2 v3 v3
Request Function
t
rf (t)
Synchronous Digraph Real-Time Tasks - 8/18 - Mohaqeqi, Abdullah, Guan and Yi
DRT Schedulability
v1 v2
v3
〈1〉 〈2〉
〈1〉
15
9
10
8
t0 5 10 15 20 25 30 35
v1 v2 v3 v3
Request Function
t
rf (t)
Synchronous Digraph Real-Time Tasks - 8/18 - Mohaqeqi, Abdullah, Guan and Yi
DRT Schedulability Condition
Notation:A set of tasks τ = {T1,T2, . . . ,Tn}πi : A path in Ti ’s graph
Theorem (Stigge 2013)A job with WCET “e” and relative deadline “d” is schedulable under aset of higher priority tasks τ if and only if for all (π1, . . . , πn) ∈ Π(τ):
∃t ≤ d : e +∑Ti∈τ
rf πi (t) ≤ t (1)
rf πi (t) could be derived independently.
Synchronous Digraph Real-Time Tasks - 9/18 - Mohaqeqi, Abdullah, Guan and Yi
DRT Schedulability Condition
Notation:A set of tasks τ = {T1,T2, . . . ,Tn}πi : A path in Ti ’s graph
Theorem (Stigge 2013)A job with WCET “e” and relative deadline “d” is schedulable under aset of higher priority tasks τ if and only if for all (π1, . . . , πn) ∈ Π(τ):
∃t ≤ d : e +∑Ti∈τ
rf πi (t) ≤ t (1)
rf πi (t) could be derived independently.
Synchronous Digraph Real-Time Tasks - 9/18 - Mohaqeqi, Abdullah, Guan and Yi
DRT Schedulability Condition
Notation:A set of tasks τ = {T1,T2, . . . ,Tn}πi : A path in Ti ’s graph
Theorem (Stigge 2013)A job with WCET “e” and relative deadline “d” is schedulable under aset of higher priority tasks τ if and only if for all (π1, . . . , πn) ∈ Π(τ):
∃t ≤ d : e +∑Ti∈τ
rf πi (t) ≤ t (1)
rf πi (t) could be derived independently.
Synchronous Digraph Real-Time Tasks - 9/18 - Mohaqeqi, Abdullah, Guan and Yi
SDRT Schedulability
v1 v2
v3
〈1〉 〈2〉
〈1〉
15
9
10
8
s1
t0 5 10 15 20 25 30 35
v1 v2 v3 v3
t
rf (t)
s1
Synchronous Digraph Real-Time Tasks - 10/18 - Mohaqeqi, Abdullah, Guan and Yi
Alignment
v1
v2
v3
v4
4
1
1
2
s1
15
40
25
2510
Task T1:
v5 v6
v7
2 1
1
s1
20
9
10
8
Task T2:
0 5 10 15 20 25 30 35 40T2
0 5 10 15 20 25 30 35 40T1
v1 v2 v3
v5 v6
Alignment
0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40
v1 v2 v3
v5 v6
0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40
v1 v2 v3
v5 v6blocked
Unsynchronized
Synchronized (Aligned)
rf 1
s1
rf 2
s1
rf 1
s1
rf 2
s1
Alignment
0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40
v1 v2 v3
v5 v6
0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40
v1 v2 v3
v5 v6blocked
Unsynchronized Synchronized (Aligned)rf 1
s1
rf 2
s1
rf 1
s1
rf 2
s1
SDRT Schedulability Condition
τ = {T1,T2, . . . ,Tn}πi : A path in Ti ’s graph
TheoremA job with WCET “e” and relative deadline “d” is schedulable under aset of tasks τ if and only if for all π = (π1, . . . , πn) ∈ Π(τ), ∀R ∈ RF π:
∃t ≤ d : e +∑
rf i∈Synch(R)Ti∈τhp
rf i (t) ≤ t
Efficient Exploration� Removing dominated request function� Search using an “abstraction and refinement” approach
Synchronous Digraph Real-Time Tasks - 13/18 - Mohaqeqi, Abdullah, Guan and Yi
SDRT Schedulability Condition
τ = {T1,T2, . . . ,Tn}πi : A path in Ti ’s graph
TheoremA job with WCET “e” and relative deadline “d” is schedulable under aset of tasks τ if and only if for all π = (π1, . . . , πn) ∈ Π(τ), ∀R ∈ RF π:
∃t ≤ d : e +∑
rf i∈Synch(R)Ti∈τhp
rf i (t) ≤ t
Efficient Exploration� Removing dominated request function� Search using an “abstraction and refinement” approach
Synchronous Digraph Real-Time Tasks - 13/18 - Mohaqeqi, Abdullah, Guan and Yi
Experiments: Analysis Efficiency
0 10 20 30 400
5
10
15
Number of Total Actions (Utilization = 0.5)
Run
-Tim
e(secon
ds)
0 10 20 30 400
5
10
15
Number of Total Actions (Utilization = 0.7)
Run
-Tim
e(secon
ds)
Experiments: Analysis Efficiency
0 10 20 30 400
5
10
15
Number of Total Actions (Utilization = 0.5)
Run
-Tim
e(secon
ds)
0 10 20 30 400
5
10
15
Number of Total Actions (Utilization = 0.7)
Run
-Tim
e(secon
ds)
v1 v2
v3
s1
s2
v4 v5
v6
s1
s2
refinementlevel
0 1 2 3 ... n
under-approx.
over-approx.
τ
Step 1
Step 2
Over-approx. v1 v2
v3
v4 v5
v6
v1 v2
v3
s1v4 v5
v6
s1
Und
er-app
rox. v1 v2
v3
v4 v5
v6
v1 v2
v3
s1v4 v5
v6
s1
Synchronous Digraph Real-Time Tasks - 15/18 - Mohaqeqi, Abdullah, Guan and Yi
v1 v2
v3
s1
s2
v4 v5
v6
s1
s2
refinementlevel
0 1 2 3 ... n
under-approx.
over-approx.
τ
Step 1 Step 2
Over-approx. v1 v2
v3
v4 v5
v6
v1 v2
v3
s1v4 v5
v6
s1
Und
er-app
rox. v1 v2
v3
v4 v5
v6
v1 v2
v3
s1v4 v5
v6
s1
Synchronous Digraph Real-Time Tasks - 15/18 - Mohaqeqi, Abdullah, Guan and Yi
Experiments
0 10 20 30 400
5
10
15
Number of Total Actions (Utilization = 0.5)
Run
-Tim
e(secon
ds) Without abstraction and refinement
With abstraction and refinement
0 10 20 30 400
5
10
15
Number of Total Actions (Utilization = 0.7)
Run
-Tim
e(secon
ds) Without abstraction and refinement
With abstraction and refinement
Outline
1 A Review on DRT
2 Synchronous DRT
3 Schedulability Analysis
4 Conclusion
Synchronous Digraph Real-Time Tasks - 17/18 - Mohaqeqi, Abdullah, Guan and Yi
Conclusion and Future Work
SDRT as an extension of DRT
Expressivenessperodicsporadi
. . . DRT SDRT Timed Automata
Multicore Scheduling• Task-level paritioning• Job-level paritioning
Synchronous Digraph Real-Time Tasks - 18/18 - Mohaqeqi, Abdullah, Guan and Yi
Conclusion and Future Work
SDRT as an extension of DRT
Expressivenessperodicsporadi
. . . DRT SDRT Timed Automata
Multicore Scheduling• Task-level paritioning• Job-level paritioning
Synchronous Digraph Real-Time Tasks - 18/18 - Mohaqeqi, Abdullah, Guan and Yi
Schedulability Analysis ofSynchronous Digraph Real-Time Tasks
Morteza Mohaqeqi, Jakaria Abdullah, Nan Guan, Wang Yi
Uppsala University
ECRTS 2016
Thanks!
Appendix
Request Function DominanceAbstraction and RefinementExperiment SettingExperiments: Path Combinations (RF Dominance)Experiments: Acceptance RatioWhy Synchronized Release?Multirate TasksCritical InstantSDRT vs. DAG
Experiment Settings
Table: Task set parameters
Task Type Small Medium Large
Vertices [3, 5] [5, 9] [7, 13]
Branching degree [1, 3] [1, 4] [1, 5]
p [50, 100] [100, 200] [200, 400]
e [1, 2] [1, 4] [1, 8]
d [25, 100] [50, 200] [100, 400]
Number of Path Combinations
Number of path combinations that should be considered inschedulability analysis
1.00E+00
1.00E+06
1.00E+12
1.00E+18
1.00E+24
1.00E+30
1.00E+36
1.00E+42
1.00E+48
1.00E+54
1.00E+60
1.00E+66
0 0.2 0.4 0.6 0.8 1
To
tal co
mbin
atio
ns
Utilization
Total combinations
SDRT Dominance (3n actions)
SDRT Dominance (n actions)
SDRT Dominance (No action)
Schedulability Analysis Results
Schedulability analysis results for different number ofsynchronizations
Acceptance Ratio Tested Combinations
Util. No act. n act. 3n act. No act. n act. 3n act.
0.35 1 1 1 37 37 37
0.4 1 1 1 52 52 52
0.45 1 1 1 70 70 70
0.5 0.94 0.96 0.96 116 165 14768
0.55 0.6 0.77 0.85 154 218 46694
0.6 0.1 0.19 0.26 225 392 59114
0.65 0 0 0.05 178 372 19167
Why Execution-Independent Synchronization?
Separation of Computationand Communication• More predictability
Ada’s Rendezvous mechanism
Fixed input/output instants
Why Execution-Independent Synchronization?
Separation of Computationand Communication• More predictability
Ada’s Rendezvous mechanism
Fixed input/output instants
SDRT Modeling Usage
Engine control tasks(Davis-2014, Biondi-2014)
Multirate controllers Rate-dependent behaviour
f1 f2
f1 f2
f3
s1
p1 p2
p3
v1 v2
v1 v2
s1p4 p5
SDRT Modeling Usage
Engine control tasks(Davis-2014, Biondi-2014)
Multirate controllers Rate-dependent behaviour
f1
f2
f1 f2
f3
s1
p1 p2
p3
v1
v2
v1 v2
s1p4 p5
SDRT Modeling Usage
Engine control tasks(Davis-2014, Biondi-2014)
Multirate controllers Rate-dependent behaviour
f1
f2f1 f2
f3
s1
p1 p2
p3
v1
v2v1 v2
s1p4 p5
Request Function Dominance
t
rf (t)
0 1 2 3 4 5 6 7 80
1
2
3
4
5
rf 2s
rf 1
s
ts t′s
t
rf (t)
0 1 2 3 4 5 6 7 80
1
2
3
4
5
rf 2s
rf 1
LemmaA request function rf 1 dominates a request function rf 2 if:
1 ∀t : rf 1(t) ≥ rf 2(t),
2 rf 1 and rf 2 contain the same sequence of actions, and
3 (ASrf1 is empty) or (ts ≤ t′s and rf 1(ts) ≥ rf 2(t′s) and rf ′1 dominates rf ′
2), where(s, ts) = ASrf 1 [0], (s, t
′s) = ASrf 2 [0], and rf ′
1 and rf ′2 are obtained by
Align_and_Pop(rf 1 , rf 2 , s).
Abstraction and Refinement
Abstraction:
Refinement:
t
rf (t)
0 1 2 3 4 5 6 7 8 9 10 11 120
1
2
3
4
5
6 rf 1rf 2
The most abstractrequest function
Concreterequest functions
Abstraction and Refinement
Abstraction:
Refinement:
t
rf (t)
0 1 2 3 4 5 6 7 8 9 10 11 120
1
2
3
4
5
6rf 1 t rf 2
rf 1rf 2
The most abstractrequest function
Concreterequest functions
Abstraction and Refinement
Abstraction:
Refinement:
t
rf (t)
0 1 2 3 4 5 6 7 8 9 10 11 120
1
2
3
4
5
6rf 1 t rf 2
rf 1rf 2
The most abstractrequest function
Concreterequest functions
Critical (Scheduling) Instant
v1 v2〈1, 1〉 〈1, 5〉
1
5
s1
T1
v3
〈1, 3〉
3
T2
J
〈1, 3〉
3, s1
T3
v1 v2
v1 v2
v3v3
v3 v3
t0 1 2 3 4
Jdeadline missof J
The critical instant for J is not
necessarily when all the tasks are
released simultaneously with J.
Future Work
Broadcast synchronizationCritical instant for the general case
References
[Stigge-2013] M. Stigge and W. Yi, “Combinatorial abstraction refinement for feasibilityanalysis,” Real-Time Systems Symposium (RTSS), 2013.
[Sun-2016] J. Sun, N. Guan, Y. Wang, Q. Deng, P. Zeng, and W. Yi, “Feasibility of fork-join real-time task graph models: hardness and algorithms,” ACM Trans. Embed.Comput. Syst. (TECS) 2016.
[Guan-2011] N. Guan, P. Ekberg, M. Stigge and W. Yi, “Resource sharing protocolsfor real-time task graph systems,” Euromicro Conference on Real-Time Systems(ECRTS), 2011.
[Biondi-2104] R. I. Davis, T. Feld, V. Pollex and F. Slomka,“Schedulability tests for taskswith variable rate-dependent behaviour under fixed priority scheduling,” Real-Timeand Embedded Technology and Applications Symposium (RTAS), 2014.
[Davis-2104] A. Biondi, A. Melani, M. Marinoni, M. D. Natale and G. Buttazzo, “Ex-act interference of adaptive variable-rate tasks under fixed-priority scheduling,”Euromicro Conference on Real-Time Systems, Madrid, 2014.