Evaluation Criteria to Run Scientific Applications in the Cloud - Cloud Computing with e-Science Applications - page 14

basis; to obtain the number of days, the yearly cost is divided by the cost
efficiency factor times 24. The following equation represents the break-even
point calculation:

BS
CE

YC

BEP =

× 24

F

where BEP represents the break-even point, BS YC represents the calculated
yearly cost of the base system, CE F represents the cost efficiency factor, and
24 is the number of hours in a day. The result of this equation is expressed
in number of days after which it becomes more cost efficient to use a server
cluster instead of a cloud. It is important to remember that the number of
days expressed by this equation is for continuous usage, 24 hours per day,
but real-world usage is normally less than that. In a practical approach, if the
server is used for fewer days per year than the break-even point, it is cheaper
to use the cloud instead.

1.8 Evaluation of Providers: A Practical Example
To provide a better understanding of the proposed methodology, we will
evaluate a hypothetical scenario. For this scenario, we need to execute the
weather forecast for a region on a daily basis; the application is already
developed in the Unix environment. Consider that we actually use a cluster
to execute the application; now, this cluster needs to be changed because the
supplier does not provide maintenance for it. We want to compare the acqui-
sition of a new cluster to a public cloud provider to verify which presents the
best solution in our case.
The first step is to verify if the application can be executed on both sys-
tems; because of the Unix execution model, it is compatible with the new
cluster and with the cloud since both have a compatible operating system.
The cloud provides adequate tools to create a cluster-like environment to
execute parallel applications, and the delivery procedures are performed
using standard network protocols, such as FTP (file transfer protocol). The
conclusion is that the application can be executed both on the new cluster
and in the cloud.
The second step is related to the performance of the solutions; it is neces-
sary to execute the same workload on both and then calculate the overhead,
in terms of execution time, of the solutions. The workload in our example is
the weather forecast application itself, with real input data, and we assume
the cluster as the base system and the cloud as the candidate system. The
execution time for the cluster was 4 hours (240 minutes), and the execution