Achim Lösch and Marco Platzner achimloesch platzner upbde Heterogeneous Compute Node Contribution Novel energyoptimizing list scheduling approach for single heterogeneous compute nodes based on real measurements ID: 626852
Download Presentation The PPT/PDF document "reMinMin : A Novel Static Energy-Centric..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
reMinMin: A Novel Static Energy-Centric List Scheduling Approach Based on Real Measurements
Achim Lösch and Marco Platzner
{
achim.loesch
,
platzner
}@upb.deSlide2
Heterogeneous Compute Node
Contribution:Novel energy-optimizing list scheduling approach for single heterogeneous compute nodes based on real measurements
2Slide3
Energy Scheduling
Related Work:Energy-minimizing list schedulers, e.g., Energy-Aware MinMin [1] and Minimum Energy-Minimum Energy [2]Do not consider energy consumed by idling resources
MINMIN [3] adjusts estimated energy consumption between a min and max value, depending on the number of cores allocated to a task
Considers energy consumed by idling resources but energy model not applicable to non-CPU architectures
Our approach:
Considers both, dynamic and static energy consumption
Energy model more precise than in related work
Feasible for CPUs and accelerators with power sensorsConsiders that tasks executed on accelerators induce
energy consumption on hostEnergy data measured on real system instead of estimations
3Slide4
Energy Model – Determining Idle Power
4
P
t
r
FPGA
P
t
r
GPU
P
t
r
CPU
T(
τ
SLEEP
,r
CPU
)
R = {
r
CPU
,
r
GPU
,
r
FPGA
}
E
total
(R|
τ
SLEEP
,r
CPU
)
P
idle(R)
①
Heterogeneous Compute Node
P
idle
(
r
CPU
)
≈
16.9 W
P
idle
(
r
GPU
)
≈
26.9 W
P
idle
(
r
FPGA
)
≈
23.8 W
P
idle
(R)
≈
67.7 WSlide5
Energy Model – Task-induced Energy
5
P
t
r
FPGA
P
t
r
GPU
P
t
r
CPU
T(
τ
i
,r
GPU
)
E
idle
(R|
τ
i
,r
GPU
) =
T(
τ
i
,r
GPU
)
·
P
idle
(R)
②
E
task
(R|
τ
i
,r
GPU
) =
E
total
(R|
τ
i
,r
GPU
) –
E
idle
(R|
τ
i
,r
GPU
)
E
total
(R|
τ
i
,r
GPU
)
③
R = {
rCPU, rGPU, rFPGA }
P
idle
(R)
①Slide6
Energy Model – Total Energy
6
P
t
r
FPGA
P
t
r
GPU
P
t
r
CPU
T(
τ
i
,r
GPU
)
①
②
E
total
(R|
τ
i
,r
GPU
) =
E
task
(R|
τ
i
,r
GPU
)
+ T(
τ
i
,r
GPU
)
·
P
idle
(R)
③
R = {
r
CPU
,
r
GPU
,
r
FPGA
}
Measure
N
tasks @
M
resources
(
N
∙
M
task-resource pairs)
Scheduler input:
① P
idle(R)
②
ETC[N][
M]
Expected Time for Completion
③
Etask[N][
M]Slide7
reMinMin Approach
repeat:
for each
task-resource pair:
Calculate
completion time (
)
Update system’s static energy consumption (
)
Calculate system’s total energy consumption (
)
endAssign task to resource with overall minimum
Remove assigned task from task
set
until
all tasks are assigned
7Slide8
Example
8
P [W]
t [s]
0
6
2
4
12
8
10
18
14
16
24
20
22
10
20
30
40
0
P [W]
t [s]
0
6
2
4
12
8
10
18
14
16
24
20
22
10
20
30
40
0
50
50
①
②
③
ETC
ETCSlide9
Example
9
0
6
2
4
12
8
10
18
14
16
24
20
22
10
20
30
40
0
0
6
2
4
12
8
10
18
14
16
24
20
22
10
20
30
40
0
50
50
1)
2)
3)
ETC
ETC
P [W]
t [s]
P [W]
t [s]Slide10
Example
10
0
6
2
4
12
8
10
18
14
16
24
20
22
10
20
30
40
0
0
6
2
4
12
8
10
18
14
16
24
20
22
10
20
30
40
0
50
50
ETC
ETC
1)
2)
3)
P [W]
t [s]
P [W]
t [s]Slide11
Example
11
0
6
2
4
12
8
10
18
14
16
24
20
22
10
20
30
40
0
0
6
2
4
12
8
10
18
14
16
24
20
22
10
20
30
40
0
50
50
Considering
idle energy is key to optimize total energy consumption.
ETC
ETC
P [W]
t [s]
P [W]
t [s]Slide12
Paper/Poster Outline
Present reMinMin in more detailExperiments
Comparison to 2 scheduling approaches
Minimum of
Optimum scheduler (exhaustive search)
3 task sets
Homogeneous task set
Heterogeneous task set
Mixed task setTask sets consist of 16 tasksInstances of 4 applications
Results from experimentsreMinMin outperforms Minimum of
approachreMinMin
even close to Optimum scheduler
12Slide13
Thank you for your attention!
13
References:
[1] Y. Li, Y. Liu, and D. Qian, “A heuristic energy-aware scheduling algorithm for heterogeneous
clusters,” in
2009 15th International Conference on Parallel and Distributed Systems
, Dec 2009
[2] J. K. Kim, H.J. Siegel, A. A. Maciejewski, and R. Eigenmann, “Dynamic resource management
in energy constrained heterogeneous computing systems using voltage scaling,” IEEE
Transactions on Parallel and Distributed Systems, vol. 19, no. 11, Nov 2008[3] S. Nesmachnow, B.
Dorronsoro, J. E. Pecero, and P. Bouvry, “Energy-aware scheduling on multi-
core heterogeneous grid computing systems,” Journal of Grid Computing, vol. 11, no. 4, 2013