Cost-effective data transfer for data creation and data recovery - Reliability Assurance of Big Data in the Cloud - page 74

t 1 and deadline are set as shown in Figure 7.8 and the bandwidth scheduling process
starts from t 1 :
1. Between t 1 and t 2 , the initial available bandwidth BW 2 is smaller than minimumBW so that
the link rate increases twice where the available bandwidth of the link increases to BW 2 9
and BW 2 99 respectively. Because BW 2 99 is larger than minimumBW and smaller than maxi-
mumBW , BW 2 99 is allocated in this time slot (striped area as shown in Figure 7.8 ).
2. TS ( t 2 ) is skipped because of shutdown period.
3. Between t 3 and t 4 , the available bandwidth of the link is already larger than minimumBW and
smaller than maximumBW . Hence, link rates remain unchanged and the current available
bandwidth is allocated.
4. The bandwidth scheduling process repeats for each timeslot. Between t n − 1 and deadline , the
available bandwidth is larger than maximumBW so maximumBW is allocated.
5. All the bandwidth is allocated in which the data transfer task is expected to be completed
between t n − 1 and deadline.

7.5 Evaluation of LRCDT

In this section we present the evaluation of our LRCDT strategy. To validate the ef-
fectiveness of the strategy in reducing energy consumption, we compare LRCDT with
two other existing popular strategies [4,78] from the aspects of energy consumption
and task completion time respectively. As mentioned in Section 2.3 , the strategy pro-
posed by Chun et al. [4] is to transfer the data file in typical lazy fashion, where data
transfer is conducted with a constant minimum speed and completes by the deadline.
Meanwhile, the strategy proposed by Hays [78] transfers the data file in a typical eager
fashion and that data transfer is conducted at the maximum speed available. Accord-
ing to the characteristics of these two strategies, we name them as the minimum-speed
strategy and the maximum-speed strategy respectively.
In the evaluation, we build an environment to simulate data transfer links of a real
Cloud network. All the three data transfer strategies are simulated as three differ-
ent bandwidth scheduling processes following different rules. Simulations of all three
strategies are conducted based on randomly generated data transfer links with random
traffic conditions. We generate multiple data transfer links with different parameters,
and conduct the bandwidth scheduling processes with different rules on each data
transfer link. We obtained the bandwidth usage of data transfer links and calculated
the overall energy consumption during the period of task execution. Each simulation
with certain parameter sets was conducted several times, and all the simulation results
are the average results of the simulations.

7.5.1 Parameters of simulation
We have simulated all the three strategies based on randomly generated data transfer
links. In order to simulate the traffic conditions of a real data transfer link in the Cloud,
in each overall agenda of the generated links, the timeslots, available bandwidth
and link rate of each router are generated based on parameters including ( startTime ,