Information Technology Reference
In-Depth Information
partial hour that a machine is working on a particular execution, the user
who launched that execution will have to pay a fixed amount. The hourly
rate varies according to machine technical specifications, namely, more
power means more money. Considering this, the formula for charging com-
putational resource usage is the following:
Total computational cost = (Total # of machine hours per execution)
* (Hourly rate depending on machine size)
6.5.4.3 Additional Costs
The e-Clouds proposal includes an additional fixed rate for data communica-
tion and queue resource usage. This includes the cost for traffic to and from
the machines and the costs associated with queue service. Every user should
pay the same fixed rate for each new execution launched on e-Clouds. This
way, the total additional cost would be calculated using the following formula:
Total additionals = (# of executions) * (Fixed rate for an execution)
There are two main reasons behind the decision for leaving these addi-
tional costs as a fixed rate. The first one is that, for most cloud providers,
communication costs are almost insignificant compared to those of stor-
age and computing. This is not true for all cases, but experience with some
scientific applications showed that it somehow follows in general. The second
reason for this management decision is that it can be complex to account for
all communication processes that happen in a machine. This complexity has
an impact not only on the e-Clouds daily operation but also on the develop-
ment of new functionalities.
6.6 Results
Several tests were created to obtain an idea of the cost and time relationship
under an AWS platform. Different instance types were used to check the per-
formance under different technical scenarios. In particular, in AWS jargon,
the following instance types were used: c1.medium (2 Cores 2,5 EC2 Units,
1.7 GB RAM, moderate I/O performance) and c1.xlarge (8 Cores 2,5 EC2
Units, 7 GB RAM). The instance selection resembles the machine specifica-
tions of a private cluster where some analyses were executed.
Maxent software is a simple .jar file that runs just like any other Java appli-
cation. It displays a graphical interface with some options so that the user can
define certain values that are relevant to the maximum entropy modeling
approach. A custom version of Maxent was used to perform the tests. It was
Search WWH ::
Custom Search