which each data file starts transmission at the beginning
of a time interval and finishes at the end of it, and routing
and scheduling for requests under a centralized network
control.
The time-expanded network technique was applied
in [6–8,12–15] to formulate networks with time-
varying link capacity into static ones, which makes the
problems tractable for optimization techniques. The
time-expanded network technique is similar to our pro-
posed framework; their differences will be elaborated
upon in Section VI. The benefits of applying assistive
storage in bulk data transfer are briefly summarized
in Table I.
III. TIME-SHIFTED MULTILAYER GRAPH
A. System Model and Assumptions
Without loss of generality, we assume the structure of an
optical switch node equipped with assistive storage shown
in Fig. 1. Data from the incoming ports can be temporarily
stored at the storage element by configuring the optical
switch. The optical switch is assumed to have sufficient
ports connected to the assistive storage to avoid read/write
conflicts while accessing the assistive storage. The assis-
tive storage element can employ different storage technol-
ogies. Some typical storage technologies available for bulk
TABLE I
A
SSISTIVE STORAGE IN BULK DATA TRANSFER APPLICATIONS AND BENEFITS
Goal Major Assumptions
Impact of SnF Versus That of
Other Approaches,
Especially E2E
a
Assistive storage used in
circuit-switched networks
Improve network throughput
[3]
Time-slotted transfer,
centralized schedule, perfect
prior knowledge of network,
limited storage
Network throughput was
improved up to 18% by SnF.
Reduce request blocking [4,5] Time-slotted transfer,
centralized schedule, perfect
prior knowledge of network,
unlimited storage
In [5], request blocking was
reduced up to 47% by SnF.
Assistive storage used in
packet switched networks
Minimize cost of bulk data
transfer [10]
No routing or schedule issues,
unlimited storage
40 TB data were transferred
for free by SnF on a daily
basis but incurred
thousands of dollars of cost
under E2E.
Minimize bulk data transfer
time [8]
Time-slotted transfer,
centralized schedule, no
prior knowledge of network,
limited storage
NetStitcher carried five times
more data than others but
took a few hours, whereas
others took days.
Minimize cost of bulk data
transfer [7,14]
Time-slotted transfer,
centralized schedule,
perfect/imperfect prior
knowledge of network,
unlimited storage
In [7], cost was reduced up to
75% by SnF.
Maximize overall bulk data
transfer utility [12]
Time-slotted transfer,
centralized schedule, perfect
prior knowledge of network,
unlimited storage
A bulk data transfer system
employing SnF was built
within an SDN architecture.
Reduce peak traffic and
balance network traffic [6]
Time-slotted transfer,
centralized schedule, perfect
prior knowledge of network,
limited storage
Temporally, peak traffic was
reduced up to 13%; spatially,
the traffic distribution was
balanced to be up to 10%
better by SnF.
Reduce interdomain traffic
[13]
Time-slotted transfer,
centralized schedule, perfect
prior knowledge of network,
unlimited storage
Interdomain traffic was
reduced up to 50%.
Minimize incurred delay, or
maximize the amount of
data delivered [
15]
Time-slotted transfer,
centralized schedule, perfect
prior knowledge of network,
limited storage
Incurred delay for the
delivery of a certain amount
of data was avoided when
the storage capacities at
intermediate nodes were
sufficient.
a
E2E refers to end-to-end data transfer approaches.
164 J. OPT. COMMUN. NETW./VOL. 8, NO. 3/MARCH 2016 Lin et al.