Database Reference
In-Depth Information
research on the Cloud with the replication-based data storage scheme, our research could
benefit the most for data-intensive applications in the Cloud.
Although the Cloud with a replication-based data storage scheme is the premise of
our research, our data storage and data reliability assurance solution is also applicable
for the Cloud with erasure coding-based data storage schemes. We will discuss this in
Chapter 8 as part of our future work.
3.2.3 Cloud network and data transfer activities
In the life cycle of Cloud data, creation and recovery of data are mainly to transfer
the replicas within the scale of the Cloud network and store them in the appropriate
location. For facilitating our research on maintaining the Cloud data, characteristics
of Cloud network and data transfer activities within the Cloud network need to be
analyzed. Specifically, for investigating the big data storage in the Cloud, analyses in
this area are considered to be important.
First, for the Cloud network, in Section 2.3 , we mentioned that Cloud systems (public
Clouds specifically) are primarily running based on data centers with dedicated networks,
interconnections with dedicated links. Although bandwidth reservation has not been com-
monly utilized in current Cloud networks yet, the dedicated feature of Cloud networks,
makes the bandwidth reservation feasible to be implemented, where the high performance
and fully controllable features of bandwidth-reserved networks are highly desirable. With
the advantages of bandwidth reservation, challenges that commonly exist in Clouds, such
as performance prediction of data transfer services and availability of data, can be tackled
effectively. Currently, many high-end network routing devices contain bandwidth control
and reservation features, hence implementing bandwidth reservation in the Cloud is practi-
cal. Therefore, our research assumes a Cloud where bandwidth reservation is enabled on its
dedicated Cloud networks. According to the preceding analysis, we consider the assumption
of bandwidth reservation ability to be reasonable.
Second, data transfer activities in the Cloud network have two major purposes, which are
data maintenance and data processing. Meanwhile, a Cloud may be composed of several
data centers, and hence the data transfer in the Cloud is conducted both within a data center
and on the links between data centers for data center interconnection (DCI). In order to
transfer data within the entire Cloud network while meeting the demands of data transfer
with different purposes, all these different types of data transfer activities need to be
investigated.
3.2.3.1 Case for data maintenance within the data center
When data are generated, replicas of the data are transferred to appropriate storage
devices within the data center [8,9] . Similarly, when a storage device fails, replicas
are recovered via copying other replicas and transferring them to new storage devices.
For these data maintenance activities, data transfer does not need to be conducted at
the highest speed so that other applications can access the data promptly if needed.
Instead, there usually exists a time limit as the upper bound for the duration of data
transfer. Failing to complete the data transfer within this time limit could cause prob-
lems, for example, violation of service level agreements [33] . Meanwhile, this type
 
Search WWH ::




Custom Search