Cost-effective data reliability assurance for data maintenance - Reliability Assurance of Big Data in the Cloud - page 58

Amazon Beanstalk, mainly for evaluating the metadata scanning procedure and the
proactive replica checking procedure. The structure of the experimental PRCR con-
sists of one user interface and one PRCR node, both of which run on a single Amazon
EC2 instance. Based on the minimum replication algorithm, execution times of the
metadata scanning process and the proactive replica checking task are obtained.
In the experiments, the metadata scanning procedure and the proactive replica
checking procedure are both simulated with several configurations. We hire four types
of Amazon EC2 compute instances for the management of 3000 S3 objects (i.e., data
files) stored with standard storage service and reduced redundancy storage service,
respectively. Table 6.2 shows the results of the experiments. It can be seen that the
metadata scanning time is at a magnitude of hundreds of nanoseconds, and the proac-
tive replica checking time is at a magnitude of tens of milliseconds.

6.5.1.3 Impact of PRCR on data access performance
Compared with the conventional three-replica strategy, by using PRCR, because of the
reduction in data replication, data access performance could potentially be affected. For
evaluation, we conducted data access speed tests with Amazon S3 so as to analyze the
impact of storing no more than two replicas compared with storing the conventional
three replicas. As Amazon states that “latency and throughput for reduced redundancy
storage are the same as for standard storage” ( http://docs.aws.amazon.com/AmazonS3/
latest/dev/UsingRRS.html ), we conducted data access performance tests for standard
Amazon S3 only. Specifically, we created AWS EC2 instances in different AWS re-
gions as well as used a local computer at Swinburne University of Technology in Mel-
bourne, Australia, to experiment with the data transfer speed for accessing files stored
in Amazon S3 in different regions. In addition to the location of the data source and the
data transfer target, the configuration of the data transfer tests were all the same.
The results are shown in Table 6.3 . Based on the results, a major observation can
be clearly seen, that is, data transfer within the same region is always of the highest
speed, where the data are transferred much quicker than that between different places
( > 3000 KB/s vs. < 300 KB/s). According to these results, in summary, we have the
following conclusions for the impact of using PRCR on data access performance com-
pared with the conventional three-replica strategy.
First, in the case that all replicas are stored in one region in practice as in Ama-
zon S3, as addressed earlier, according to the description of Amazon S3, latency and

Table 6.2 Metadata scanning time and proactive replica checking
time

t1.micro m1.small m1.large m1.xlarge

Scanning time ≈700 ns ≈400 ns ≈700 ns ≈850 ns

checking time (standard) ≈27 ms ≈27 ms ≈30 ms ≈27 ms

checking time (reduce
redundancy)

≈25 ms ≈24 ms ≈37 ms ≈23 ms