Barbara Pernici, Politecnico di Milano, Italy Characterizing Energy

Characterizing energy consumption and
adaptation strategies for
Cloud Applications
Barbara Pernici
Dipartimento di Elettronica, Informazione e Bioingegneria
Politecnico di Milano
[email protected]
Outline
• An approach towards an adaptive behaviour for improving energy
efficiency and reducing CO2 emissions in federated clouds
• The ECO2Clouds project
• Analyzing batch workloads
Barbara Pernici - ACROSS, September 2016
2
Improving energy efficiency in data centers and decrease CO2
emissions
Key concepts:
• Energy
efficiency
• Green
indicators
• Monitoring
• Adaptation
actions
• Multilayer
European projects: GAMES, ECO2Clouds
http://eco2clouds.eu
Barbara Pernici - ACROSS, September 2016
3
Barbara Pernici - ACROSS, September 2016
4
Multilayer
5
Eco-metrics
Barbara Pernici - ACROSS, September 2016
6
Measuring power consumption
7
Monitoring infrastructure
Barbara Pernici - ACROSS, September 2016
8
Adaptation strategies for applications
Barbara Pernici - ACROSS, September 2016
9
Application profiles
Barbara Pernici - ACROSS, September 2016
10
Strategies
Barbara Pernici - ACROSS, September 2016
11
Designing application profiles
• Batch applications
• Modeling energy consumption in different configurations
• Meeting requirements and evaluating alternative configurations
• Refining application profiles
Barbara Pernici - ACROSS, September 2016
12
Power and energy modeling
Barbara Pernici - ACROSS, September 2016
13
Analyzing batch applications
– Power consumption of an experiment involves N VMs
– Energy per job (J jobs)
Power models
• Power consumption of physical host and VMs proportional to
usage
• Power consumption of a VM
15
Testing the model
16
Configurations
• Number of VMs for J jobs?
• Which are the parameters for the alternatives?
– Order of execution
– Shared resources: storage access before computing
activities
• What is interference between applications
• What if heterogeneous hosts are assigned by the cloud
scheduler?
17
18
Analysis of alternative configurations
• Asynchronous/synchronous
• No. of VMs (1 to J)
asynchronous
synchronous
19
Simulation models
asynchronous
synchronous
Validation – simulation with JMT vs experimental data
Synchronous accesses – 1 host
1 VM
N=J
20
Analysis of energy per job
21
Bottlenecks
Barbara Pernici - ACROSS, September 2016
22
Energy consumption analysis using the model
varying the number of VMs – homogeneous hosts
90
80
25 VMs
70
Energy per job
60
50
25 VMs
E/J asynch (Wh)
40
E/j synch (Wh)
30
1 VM
20
10
0
0.00
5.00
10.00
15.00
20.00
25.00
30.00
35.00
40.00
Execu/ on / me
Note: minimum P_idle, considering max number of VMs on hosts
23
24
In case of heterogeneous hosts- asynchronous
(2 classes – high/low; 50 jobs)
53
48
43
24 VMs
EpJ
38
33
4 VMs
28
23
18
a=0
a = 0.25
a = 0.5
a = 0.75
a=1
13
8
3
4
5
6
7
8
9
10
R time
Execution
α = % of high performance hosts
High performance: 20% faster, 20 vs 8 cores, idle power 40% of lower perf.
In case of heterogeneous machines - synchronous
(2 classes – high/low; 50 jobs)
53
25
24 VMs
48
43
EpJ
38
33
4 VMs
28
23
18
13
8
3
4
5
6
7
ExecutionRtime
α = % of high performance hosts
8
9
10
a=0
a = 0.25
a = 0.5
a = 0.75
a=1
Future work
26
• Analyze other types of applications
– Transactional
– I/O intensive
• Define algorithms to identify the best configurations for a mix
of applications
• Study other eco-metrics and dependencies between metrics
References
• U. Wajid, C. Cappiello, P. Plebani, B. Pernici, N. Mehandjiev M. Vitali,
M. Gienger, K. Kavoussanakis, D. Margery, D. Garcia Perez, P.
Sampaio, On Achieving Energy Efficiency and Reducing CO2 Footprint
in Cloud Computing, IEEE Transactions on Cloud Computing, Vol.
4(2), pp. 138-151, April 2016
• C. Cappiello, N. Ho, B. Pernici, P. Plebani, M. Vitali, CO2-aware
Adaptation Strategies for Cloud Applications, IEEE Transactions on
Cloud Computing, Vol. 4(2), pp. 152-165, April 2016
• M. Gribaudo, T.T.N. Ho, B. Pernici, G. Serazzi, Analysing the influence
of application deployment to energy consumption, E2DC,
Cambridge, LNCS, June 2014
• M. Gribaudo, B. Pernici and T.T.N. Ho, Characterizing Energy per Job
in Cloud Applications, submitted
Barbara Pernici - ACROSS, September 2016
27
QUESTIONS?
Barbara Pernici - ACROSS, September 2016
28
Characterizing energy consumption and
adaptation strategies for
Cloud Applications
Barbara Pernici
Dipartimento di Elettronica, Informazione e Bioingegneria
Politecnico di Milano
[email protected]