Cost-effective data transfer for data creation and data recovery - Reliability Assurance of Big Data in the Cloud - page 71

transfer. The eager data transfer requires the maximum transfer speed while the lazy
data transfer does not. By dividing data transfer into two types, LRCDT schedules
link bandwidth respectively to improve energy consumption while meeting the data
transfer speed requirement at the same time. It is able to significantly reduce energy
consumption specifically for data transfer tasks that do not require the maximum
transfer speed, referred to as lazy data transfer, so that the overall energy efficient
data transfer goal can be achieved. Compared to the data transfer strategies men-
tioned in Section 2.3 , first, LRCDT provides a much faster transfer speed in compar-
ison to the minimum-speed strategy as proposed by Chun et al. [4] . Second, LRCDT
only schedules the bandwidth on active links (i.e., the period when the link is ac-
tive) so that the shutdown approach is still allowed on the same link when LRCDT
is already applied. Third, LRCDT consumes much less energy during data transfer
in comparison to both the minimum-speed strategy and the maximum-speed data
transfer strategy proposed by Hays [78] . Fourth, unlike Adaptive Link Rate (ALR),
which monitors bandwidth usage and changes link rate afterward, LRCDT schedules
bandwidth before data transfer is conducted. It fully utilizes the advantages of the
dedicated network of the Cloud so that data transfer can be fully controlled. Mean-
while, LRCDT divides data transfer tasks into two types, according to the transfer
speed requirement, so that the energy consumption can be improved while the data
transfer speed requirement can be met at the same time.
In LRCDT several features have been designed accordingly for meeting the needs
of energy-efficient data transfer as well as addressing the considerations illustrated in
Section 3.2 .
• First, in order to reduce energy consumption, the basic idea of LRCDT is to limit the rout-
ers' link rates to the minimum level available. This ensures that the power consumption of
the routers is minimized. Meanwhile, by providing as much available bandwidth as possible
(without changing the link rate), the data file can be delivered as fast as possible.
• Second, to address the first issue raised in Section 3.2 for cost-effective data transfer strat-
egy, in LRCDT, a ( startTime, deadline ) pair is set for each data transfer task. Within a
bandwidth-reserved network, the ( startTime, deadline ) pair indicates the expected period
of link occupation, which is crucial to the bandwidth scheduling process. However, if these
parameters are not provided by the application, a default deadline value could be set. Ac-
cording to the size of the data file to be transferred in each task, LRCDT allocates sufficient
bandwidth within the ( startTime, deadline ) period to ensure that the task can be completed
in time. In addition, considering the medium that receives the data file, an upper boundary
maximumBW for the data transfer bandwidth is set for each data transfer task.
• Third, to address the second issue raised in Section 3.2 , in LRCDT, for lazy data transfer, the
energy-efficient data transfer is conducted where the link rate is minimized. For eager data
transfer the data file is transferred as quickly as possible while LRCDT schedules the maxi-
mum bandwidth for data transfer without considering the link rate. By conducting these two
different types of data transfer, LRCDT is able to meet the requirements of both types of data
transfer in the Cloud. To avoid affecting the existing link traffic on the Cloud network, in
LRCDT, the bandwidth is allocated based on the existing agenda of the link. No data transfer
bandwidth is allocated during the already scheduled shutdown period unless the data transfer
task cannot be completed within the maximum transfer duration and can be completed if the
shutdown period is occupied.