Literature review - Reliability Assurance of Big Data in the Cloud - page 16

hence data transfer is considered to be the main procedure during the data recovery
process. In order to recover Cloud data in a cost-effective fashion, in this section we
focus on data transfer approaches for distributed systems. In addition to data recovery,
data transfer is also intensively involved in creating replicas in the Cloud. Therefore,
reviews conducted in this section could also benefit our research for creating replicas
in the data creation stage.
Data transfer has been considered a very important research issue in the field of
high-performance networks and distributed storage systems for a long time [69,70] . In
recent years, the ever-developing Cloud and large-scale distributed storage technolo-
gies have resulted in higher demand for data transfer from both data transfer speed and
energy consumption aspects. Balancing the trade-off between data transfer speed and
energy consumption is a significant challenge.
On one hand, to meet the requirements of the large-scale data-intensive applica-
tions, the need for high-speed yet predictable data transfer is increasing where net-
works with effective bandwidth controls are required. Because of its fully controlled
feature, dedicated networks with bandwidth reservation have drawn more and more
attention. Typical examples of dedicated networks include research networks such
as the National LambdaRail [71] and the Internet2 Network [72] . In one study [73] ,
a bandwidth reservation approach via a centralized resource management platform
was proposed for providing predictable performance in research networks. The cen-
tralized management pattern has, however, limited scalability and hence constrains
the applicability of this approach. In another study [74] , a distributed bandwidth
reservation approach for reducing energy consumption in dedicated networks was
proposed, which could greatly improve the scalability issue compared to the former
proposal [73] .
On the other hand, the energy consumption for high-speed large-scale data trans-
fer is high. This has become one of the major factors that need to be considered in
large-scale storage systems. In recent years, many efforts have been made to reduce
the energy consumption incurred in large-scale data transfer. For example, a standard
was developed for defining management parameters and protocols in energy-efficient
Ethernet networks [75] . Energy consumption models [74,76] have been proposed for
switches and general network devices respectively. To reduce the energy consumption
over network links, several approaches are offered. In one approach [4] , a replica cre-
ation and recovery strategy was proposed where data transfer would be conducted with
a constant minimum speed to maintain a certain number of replicas. In other proposals
[77,78] , energy management approaches, referred to as “shutdown approaches,” are
offered. In these approaches, devices on the link are shut down when network traffic is
too low so that the energy consumption of routers and network links can be reduced.
Specifically, in one model [78] , the shutdown approach is conducted in such a way
that data are transmitted as fast as possible and the data transfer link is “idled” after
data transfer is finished. However, there could be problems for such approaches as
some other tasks might also use the same data transfer link meaning it cannot be shut
down. Different from the shutdown approaches that shut down devices to save power,
a phenomenon was observed that less energy was consumed by network devices when
operating at lower link rates [79] . One study [74] found that the power of a network