Upload
evita
View
24
Download
0
Embed Size (px)
DESCRIPTION
Energy-Aware Modeling and Scheduling of Real-Time Tasks for Dynamic Voltage Scaling. Xiliang Zhong and Cheng-Zhong Xu Dept. of Electrical & Computer Engg. Wayne State University Detroit, Michigan http://www.cic.eng.wayne.edu. Outline. Introduction and Related Work - PowerPoint PPT Presentation
Citation preview
Energy-Aware Modeling and Scheduling of Real-Time
Tasks for Dynamic Voltage Scaling
Xiliang Zhong and Cheng-Zhong Xu
Dept. of Electrical & Computer Engg. Wayne State University
Detroit, Michiganhttp://www.cic.eng.wayne.edu
2
Outline
Introduction and Related Work A Filtering Model for DVS Time-invariant Scaling Time-variant Scaling Statistical Deadline Guarantee Evaluation Conclusion
3
Motivation Mobile/Embedded devices power critical Energy-Performance tradeoff
Processor speed designed for peak performance Slowdown the processor when not fully utilized (DVS)
Challenges Maximize energy saving while providing deadline guarantee Real-time tasks could be periodic/aperiodic w/ highly
variable execution time Aperiodic tasks have irregular release times, which calls for
online decision making
4
Related Work Intensive studies for periodical tasks Algorithms for aperiodic tasks
Offline (Yao et al’95, Quan & Hu’01) Online: all timing information known only after job releases
Soft real-time: improve responsiveness (Aydin & Yang’04) Occasionally uncontrollable deadline misses (Sinha & Chakrabarty’01) Hard real-time w/complex admission control (Hong et al ’98) Maximize energy saving w/ frequency scaling(Qadi et al ’03, DVSST) On-line slack management for a general input (Lee & Shin ’04,OLDVS)
Objectives of this paper: Hard/statistical deadline guarantee for general input w/o
assumptions of task periodicity Unified, online solutions for both WCET based scheduling and
slack management
5
Task Model Independent tasks, preemptive w/ dynamic priorities Job releases (requests) to system are characterized
by a compound process in a discrete time domain: wi
(t) is the size (WCET) of ith jobs arrived during time [t-1,t)
n(t) stands for number of jobs arrived, each w/deadline td
0 1
w1(1) w1(2) w2(3) …
2 time
Input arrivals
6
System Model
Processor Model Support a continuous range of speed levels
Energy Model t: scheduling time slot, f(t): speed at time [t, t+1) l(t): load, #cycle allocated to all jobs during [t, t+1) P(l(t)): power as a function of load E(S): energy consumed according to a schedule S
7
A Filtering Model of Speed Scaling
Allocation function denotes the # cycles allocated to one job wi(t) during [t, t+1)
Decomposition of allocation function
g(), the impact of job sizes (WCETs) on scheduling h(), scaling function s(), the load’ feedback to scheduling
g h s Outputload
JobArrivals
Request Size
8
A Filtering Model (cont.)
Each job should be finished in td time g(wi(t))=wi(t)
Load Function l(t) is a sum of allocation to all jobs
Non-adaptive to load s(l(t)) = 1
9
A Filtering Model (cont.) The load function becomes a convolution of compounded
input request process and scaling function,
Scaling function h(t): Portion of resource allocated at each scheduling epoch from the arrival time ts to finish time ts+td
Design of scaling algorithm in a fitlerng system
10
Time-Invariant Scheduling
The optimal policy is to find an allocation
where
Treat h(t) as a time-invariant scaling function
The optimality is determined by the covariance matrix Ω of the input process w(t) in the order of deadline td
The optimization has a unique, closed form solution
11
Example Solutions with Different Input
Two multimedia traffic patterns (Krunz’00) Shifted Exponential
Scene-length Distribution (ACFExp)
Subgeometric scene-length distribution (ACFSubgeo)
Fractional Gaussian Noise (FGN) process with Hurst para. H=0.89
Simpsons MPEG Video Trace of 20,000 frames
Auto-Correlations of Traffic
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 50 100 150 200 250 300
Lag
ACF
ACFSUBGEO
ACFEXPON
ACFMPEG
ACFFGN
12
Example Solution (td=10)
Optimal Scaling Function
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0 1 2 3 4 5 6 7 8 9
time
h(t)
ACFSUBGEO
ACFEXPON
ACFFGN
ACFMPEG
Uniform Distributed
Higher degree of input autocorrelation has a more convexed scaling function
The uniform distributed allocation is a generalization of several existing algorithms for Periodic tasks Sporadic tasks Aperiodic tasks
13
Time-Variant Scaling Energy consumption can be reduced if the
scaling function h(t) is adaptive in response to change of input load
Make td runnable queues. Jobs with deadline j are put to queue j
Dis
patc
her Running
queue 1
+
…
Running queue 2
Running queue td
Outputload l(t)
Inputjobs
l1(t)
l2(t)
ltd(t)
14
Time-Variant Scaling Minimize energy consumption is to
Resource cap of queue j at time t
subject to:
The optimization has a unique solution:
where qj(t) is the backlog of queue j at time t.
Committed resource for jobs in queue j at time t
15
Determine cap of queue 5 at time 0: S5(0)
Illustration
Distribute the job as late as possible
load
The job is distributed to early slots as its size increases
L(t)=0
12 9 7
0 1 2 3 4 5 time slot
5
S5(0)=11
First determine current committed resource
16
Example Solution for a Sporadic TaskInput J(WCET): J1(1) released at 0, 5, J2(2) at 1, 7, J3 (1) at 3, 9. Deadline of all jobs: 4. Ji,j: jth instance of task i1. Schduling using EDF w/o scaling
1
1
0 t
0.5
3 954 7
Normalized speed
J1,1 J2,1 J3,1J1,2 J2,2 J3,2
6 10
2. Schduling using the Time Variant Scaling
(a)
0.5
0.25 J1,1
5 t1 3 7 139 11
Normalized speed
(b)
0.690.50.25
J1,1
J2,1
5 t1 3 7 139 11
(c)
0.50.25
J1,1
J2,1
0.69
J3,1
5 t1 3 7 139 11
0.69
(d)
0.5
0.25 J1,1
J2,1J3,1 J1,2 J2,2 J3,2
0.75
5 t1 3 7 139 11
Using a square energy function: 35% more energy saving compared to EDF. 8% to DVSST
17
Statistical Deadline Guarantee Worst case scenario schedulability test
Conservative pi: minimum interarrival
fmax
1
F(x)
v
worst case f
cumulative probability
Statistical guarantee Overload probability v=prob(l(t) > fmax)
18
Statistical Deadline Guarantee (cont.) Load tail distribution
A general bound w/ load mean and variance
Tight bounds based on load distribution Exact output distribution if input distribution known Estimate output distribution using a histogram
fmaxf’max
1
F(x)v
cumulative probability
b1 b2 bmaxbr-1 bmin
19
Evaluation Objectives
Effectiveness in energy savings Effectiveness of the deadline miss bound
Scheduling based on WCET No-DVS: run jobs with the maximum speed. Offline: Offline optimal algorithm of Yao:95 et a. DVSST: On-line algorithm for sporadic tasks Qadi:RTSS03 et al. TimeInvar: Time-invariant voltage scaling. TimeVar: Time-variant voltage scaling.
On-line slack management DVSST+CC (Cycle-conserving EDF): Worst case schedule using
DVSST with the reclaiming algorithm of Pillai and Shin (SOSP01). TimeVar+OLDVS: The time-variant voltage scaling and the
reclaiming algorithm of Lee and Shin:RTSS2004. TimeVar+TimeVar: A unified solution.
20
Energy consumption with the Robotic Highway Safety Marker application; A scenario in which robot keeps moving
Energy Savings
00.10.20.30.40.50.60.70.8
1 2 4 6 8 10
Path length of robot move
En
erg
y c
on
sum
pti
on
TimeVar is energy-efficient, close to Offline (5%); 7-11% better than DVSST
DVSST OfflineTimeVariant
11%
21
#tasks=30; Interarrival ~ exp(50 ms)WCET ~ n(100, 10)K
Workload variation characterized by actual execution time over worst case (BCET/WCET)
Energy Savings w/ Workload Variation
0
0.1
0.2
0.3
0.4
0.5
0.6
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
BCET/WCET ratio
En
erg
y C
on
sum
pti
on
DVSST+CC TimeVar+TimeVar Offline
TimeVariant adapts with workload variation effectively
22
Computation Speed Configuration
Computation requirement based on a general bound is better than worst case with mean interarrivals > 60 ms Tight bounds reduce the computation speed in half as interarrivals > 40 ms
Mean Interarrival time (ms)
Req
uir
ed s
pee
d (
MH
z)
Target deadline guarantee: 99%
23
Statistical Deadline Guarantee
No deadline misses under bound derived based on a general input: 100MHz
Statistics of TimeVar/TimeInvar under a tight bound: 40MHz Overload handling: reject new jobs or serve unfinished jobs in a
best-effort mode Target deadline guarantee 99%
SchedulingTimeInvariant TimeVariant
Reject Besteffort Reject Besteffort
Load mean (106) 21.6 21.7 21.6 21.74
Load var 166.7 168.9 141.3 142.8
Time mean 10.1 10.09 10.4 10.07
Time variance 8.7 8.4 8.4 8.8
Overload/Deadline misses
0.63% 0.63% 0.61% 0.61%
Deadline miss rate is effectively bounded
24
Conclusion Voltage/Speed scaling for a general task model A Filtering Model for DVS Two online policies to minimize energy usage
Time-invariant : A generalization of several existing approaches
Time-variant : Optimal in the sense it is online w/o future task timing information. Also effective for on-line slack management
Statistical deadline guarantee based on computation speed configuration.
Future work System-wide energy savings, e.g., wireless communication
and its interaction with CPU
25
Energy-Aware Modeling and Scheduling of Real-Time Tasks for
Dynamic Voltage Scaling
Thank you!