HBC
REC
EIC43,^200
86,^400
129,^600
172,^800
216,^000
259,^200
302,^400
345,^600Task time (s)×10^8Product(costs×time)
1.0
0.8
0.6
0.4
0.2
0.0Figure 22: Comparison of combined metrics (product of total
execution time and cost).
×10^5Time (s)8.0
6.0
4.0
2.0
0.043,^200
86,^400
129,^600
172,^800
216,^000
259,^200
302,^400
345,^600Task time (s)
(a) Total task execution time43,^200
86,^400
129,^600
172,^800
216,^000
259,^200
302,^400
345,^600Task time (s)
Lower BB
Middle BBUpper BB
EICCosts ($)30
25
20
15
10
5
0(b) Total costsFigure 23: Comparison of actual EIC outputs (execution time and
cost) and estimations according to the task time.
average by 65.36% and 68.51% when compared to HBC and
REC, respectively.
Figure 23shows the estimation accuracies according to
the task time. The actual execution time and cost are located
betweenthelowerandupperboundsoftheBollingerBand.
Figures23(a)and23(b)provethattheactualexecutiontime
and cost are close to the middle point of the Bollinger Band.
They state that the EIC would be able to offer approximate
ranges of total costs and task execution time to Cloud users.
Overall, the EIC significantly reduces the number of
checkpoint trials compared to the existing checkpointing
schemes. Furthermore, the rollback time is much lesser
because the EIC adaptively performs the checkpointing oper-
ation according to the execution time and price. Simulation
results showed that our scheme achieved the cost-efficiency
by reducing rollback time regardless of the resource types of
spot instances.
Analyzing history to compute the estimated interval
requires overheads such as CPU time. However, computa-
tions only involve failure probability, execution time and cost
estimations, and a range of the Bollinger Band. Considering
the advancement of modern computers, we strongly believe
it would take the minimal amount of overheads for computa-
tions.6. Conclusion
In this paper, we proposed the estimated interval-based
checkpointing (EIC) in the unreliable cloud computing envi-
ronment. The weighted moving average estimates the execu-
tion time and cost using the price history of spot instances
to improve the performance and stability of task processing.
The EIC performs the checkpointing operation, based on
price and time thresholds. The thresholds are determined
basedonthemovingaverageandthefailureprobability.
They are used to determine the checkpointing position to
recover from the potential failures of spot instances arising
from the price fluctuation. The Bollinger Bands determines
thelowerandupperboundsoftheestimatedexecutiontime
and cost. The ranges are informed to users as guidance for
their decision. The simulation results reveal that, compared
to the hour-boundary checkpointing (HBC) and rising edge-
driven checkpointing (REC), the EIC reduces the number of
checkpoints by 35.97% and 37.92%, respectively, on average
according to the user bid. It also reduces the rollback time
by72.46%and88.49%onaverage.Consequently,thetask
execution time is decreased with ETC by 35.53% over HBC
and 40.40% over REC. The EIC also provides the benefit of
the cost reduction by 36.26% over HBC and 38.52% over REC,
on average.Conflict of Interests
The authors declare that there is no conflict of interests
regarding the publication of this paper.Acknowledgment
This work was supported by the National Research Founda-
tion of Korea (NRF) Grant funded by the Korea government
(MEST) (NRF-2012R1A2A2A 02046684).References
[1] R. Buyya, C. S. Yeo, and S. Venugopal, “Market-oriented
cloud computing: vision, hype, and reality for delivering IT
services as computing utilities,” inProceedings of the 10th IEEE
International Conference on High Performance Computing and
Communications (HPCC ’08), pp. 5–13, September 2008.
[2]I.Foster,Y.Zhao,I.Raicu,andS.Lu,“Cloudcomputingand
grid computing 360-degree compared,” inProceedings of the
Grid Computing Environments Workshop (GCE ’08),pp.1–10,
November 2008.