Qing Gu, Patricia Lago and Simone Potenza (VU University, The Netherlands)
Presented by: Giuseppe Procaccianti
In 2013 IEEE 7th International Symposium on the Maintenance and Evolution of Service-Oriented and Cloud-Based Systems (MESOCA '13)
2. Up in the Clouds
Delegating Data Management to the Cloud
Cloud Migration is said
to bring many benefits.
3. Up in the Clouds
Delegating Data Management to the Cloud
Energy Efficiency
is one of the less
investigated.
4. Up in the Clouds
Delegating Data Management to the Cloud
Data as an asset
is often a candidate
for cloud migration.
5. Problem statement
Delegating Data Management to the Cloud
When is it convenient for a company to delegate data
management to the Cloud?
6. Approach
Delegating Data Management to the Cloud
● Literature Study on data classification schemas
● 55 schemas found
● 17 related to data archiving
* Position criterion considered irrelevant for the purpose of this presentation
● 3* criteria identified
7. Data Classification Criterion: Retention
Delegating Data Management to the Cloud
Definition: 2 variables, 1 constant
Long Retention time (LR) if:
retainment_period
is_retainment_set
retainment_threshold
is_retainment_set = TRUE
AND
retainment_period > retainment_threshold
8. Data Classification Criterion: Frequency of usage
Delegating Data Management to the Cloud
Definition: 1 variable, 1 constant
High frequency access (H) if:
access_frequency_index [1]
frequency_threshold
access_frequency_index > frequency_threshold
[1] Chen, Ying. "Information valuation for information lifecycle management."
Autonomic Computing, 2005. ICAC 2005. Proceedings. Second International
Conference on. IEEE, 2005.
9. Data Classification Criterion: Modification
Delegating Data Management to the Cloud
Definition: 1 variable, 1 constant
Young file (Y) if:
last_modification_period
last_modified_period_threshold
last_modification_period > last_modified_period_threshold
10. Data Classification Schema
Frequency of
Usage
Modification Retention Class
H Y LR Hot
H Y SR Hot
H O LR Cold
H O SR Cold
L Y LR Cold
L Y SR Cold
L O LR Cold
L O SR Frozen
Delegating Data Management to the Cloud
11. Value of Energy
Delegating Data Management to the Cloud
● Energy consumption for storage devices:
○ idle vs. active: no significant
difference
○ full vs. empty: no significant difference
● Solution: Value of Energy framework
○ relate energy consumption with data
value
○ find energy wastes due to storage of
valueless data
12. Value of Energy: metrics
Delegating Data Management to the Cloud
Metric Description Unit
WpGB
IdlePower / raw_size
Amount of power needed to keep
available a GB of storage space
W/GB
EffectivePower
Hot*WpGB * PUE
Amount of power used to store hot
data
Watt (W)
WastedPower
(Frozen+Cold) *WpGB*PUE
Amount of power used to store
cold/frozen data
Watt (W)
PotentiallyEffectivePower
FreeSpace * WpGB * PUE
Power used to provide free storage
space that can be used for hot data
Watt (W)
ServiceLevelPower
raw_size*capacity *WpGB*PUE
Power used for RAID and data
redundancy in general
Watt (W)
13. Industrial Case Study
Delegating Data Management to the Cloud
● Partner: Dutch telecommunication provider
● Motivation: demonstrate economic/environmental
benefits of Cloud archiving service
14. Industrial Case Study /2
Delegating Data Management to the Cloud
Hot data size 980 GB
Cold data size 2640 GB
Frozen data size 5280 GB
● Test storage system:
○ 24 disks
○ Raw space: 14.400 GB
○ Usable capacity: 13.200 GB
○ Transfer Rate: 12000 MB/s
○ PUE: 1.5
15. Industrial Case Study /3
Delegating Data Management to the Cloud
Before Migration After Migration
EffectivePower 92.61 W 92.61 W
WastedPower 748.44 W 0 W
PotentiallyEffectivePower 406.35 W 1154.79 W
ServiceLevelPower 113.4 W 113.4 W
16. Industrial Case Study /4
Delegating Data Management to the Cloud
● Too much free space may result in energy wastes as
well
● This waste can be quantified
using data growth metrics
● Assuming a growth rate of 30% for all data classes:
Future Hot data size 1656 GB
Future Cold data size 4462 GB
Future Frozen data size 8923 GB
17. Industrial Case Study /5
Delegating Data Management to the Cloud
Before Migration After Migration
EffectivePower 93 W 93 W
WastedPower 518 W 1034 W
PotentiallyEffectivePower 580 W 64 W
ServiceLevelPower 170 W 170W
18. Conclusions
Delegating Data Management to the Cloud
We evaluate data cloud
migration from the
Customer perspective.
Migration to the Cloud is
beneficial….
...under certain circumstances!
20. For the future...
Delegating Data Management to the Cloud
● More precise energy calculations
● Extend data classification criteria for different contexts [2]
● Investigate environmental and economic impacts
[2]
Masanet, E., Shehabi, A., Ramakrishnan, L., Liang, J., Ma, X.,Walker, B., H
endrix, V., and P. Mantha (2013 ). The Energy Efficiency Potential of
Cloud-Based Software: A U.S. Case Study. Lawrence Berkeley National
Laboratory, Berkeley, California.