Cost-effective data transfer for data creation and data recovery - Reliability Assurance of Big Data in the Cloud - page 72

According to the Cloud network model, all agendas of the routers on the link can be
merged into an overall agenda. Based on the overall agenda, the bandwidth scheduling
is conducted. According to Zhang et al. [80] , the link rate switching time ranges from
10 to 100 ms. This is quite considerable for high-performance data transfers. There-
fore, to eliminate additional link rate switching caused by LRCDT, the bandwidth
scheduling is conducted with a timeslot as the minimum schedule unit. In this way,
the link rate switching caused by LRCDT could be done where the link rate is already
planned to switch. In LRCDT, the bandwidth scheduling follows a simple “lower
boundary policy.” At each non- shutdown timeslot, the scheduled bandwidth should not
be smaller than a certain lower-boundary called “ minimumBW ” unless the maximum
available bandwidth of the link is smaller than it. minimumBW is the lowest average
bandwidth for ensuring the conduct of the data transfer task. It can be obtained from
the equation minimumBW= dataSize/maximumTransferDuration. This policy aims to
ensure that the data transfer task will be completed before the deadline. If the available
bandwidth of the link is smaller than minimumBW , the link rates of routers increase
to provide more available bandwidth. The order of link rate increment is according
to the available bandwidth of each router. The router that has the minimum available
bandwidth increases its link rate first so that the available bandwidth of the entire link
increases. The link rate increment stops when the available bandwidth is larger than,
or equal to, minimumBW . Afterward, the smaller one between maximumBW and the
maximum available bandwidth under the current link rate is scheduled for the data
transfer task. The bandwidth scheduling is conducted in chronological order of the
agenda and the shutdown period is avoided.
As the major part of the LRCDT strategy, Figure 7.7 shows the pseudo code of
the bandwidth scheduling algorithm for lazy data transfer. The bandwidth schedul-
ing starts with the initialization of several parameters: first (line 2), the ( startTime,
deadline ) pair is set and the maximum data transfer duration is initialized to be the
time between deadline and startTime minus the shutdown period. Second (line 3),
according to the maximum data transfer duration and size of the data file, the band-
width lower boundary minimumBW can be calculated. Third (line 4), for initializing
list TS , the timeslots of all agendas during the data transfer process are obtained ac-
cording to event list L of the overall agenda. After the initialization part, in the main
part of the algorithm (lines 5-13), it allocates bandwidth for each timeslot between
startTime and deadline : First, the shutdown period is skipped; second, if available
bandwidth of the link is smaller than the lower boundary minimumBW and still has not
reached the maximum available bandwidth of the link, the algorithm repeats the pro-
cess of link rate increase for the router with the smallest available bandwidth (i.e., the
bottleneck router that constrains the available bandwidth of the link). The repeating
process finishes when available bandwidth of the link is bigger than minimumBW or
reaches the maximum level. Third, if the available bandwidth of the link is larger than
maximumBW , only maximumBW bandwidth is allocated to the timeslot. Otherwise all
available bandwidth is allocated to the timeslot. Because of the bandwidth scheduling
lower boundary policy, in a very rare case, the algorithm cannot allocate sufficient
bandwidth for completing the data transfer task. In this case, the algorithm sets a loop
(lines 2-14) where, if there is still available bandwidth that can be allocated between