Image Processing Reference
In-Depth Information
management and a new protocol to efficiently track congestion information on
the bottleneck links inside a network without undue signaling load. Nonetheless,
to the best of our knowledge, there exists no solution that leverages these advanced
over-reservation techniques jointly with cooperative communication strategies in
RAN to improve the end-to-end energy efficiency in the network, so further
research was deemed necessary.
8.3.2 Proof-of-Concept for QoS Control Signaling Overhead
Reduction
As a proof-of-concept for the superiority of resource over-reservation over per-flow
approaches in terms of significant reduction of the signaling overhead to achieve
scalability, a simulation is performed using the network topology presented in
Fig.
8.1
and the Network Simulator (ns-2) [
91
]. For the sake of simplicity, each
network interface is configured with a capacity of 1 Gbps and 4 CoSs: one control
CoS (for control packets), one Expedited Forwarding (EF), one Assured
Forwarding (AF) and one Best-Effort (BE) [
92
], under the Weighted Fair Queuing
(WFQ) scheduling discipline [
93
]. In order to simulate a cooperative communica-
tion environment with dynamic changes of access points and therefore the paths,
20,000 session requests belonging to three different traffic types, such as CBR,
Pareto and Exponential are randomly generated and mapped to various CoSs and
ER1 and ER2 based on Poisson processes. The traffic bandwidth requests are
generated using uniform distribution between 128 Kbps and 8 Mbps. To show
more stable results, we run the simulation five times with different seeds of random
mapping of requests to CoSs. Then, the mean values are plotted for all seeds with a
confidence interval of 95 %. Further details on the simulation are available in [
94
].
Figure
8.9
shows that the resource probing and the number of reservation events
overlap when the network is less congested (request number below 4,000). As the
request number increases further, the number of probing events goes higher than
that of the reservation. This means that requests are denied when the probing
reveals insufficient resource to guarantee the QoS demanded. The release messages
events were also tracked upon session termination and the overall per-flow signal-
ling events number (probing + reservation + release) is plotted in Fig.
8.9
. Besides
the per-flow results, the resource over-reservation performance is also shown. As
one can observe, the over-reservation control effectively allows drastic reduction of
the QoS control signaling events. More importantly, no QoS signaling is triggered
when the network is less congested since the resource is reserved in advance. The
signaling is invoked only when the over-reserved resource parameters need
re-adjustment to prevent CoS starvation. Generally, in Fig.
8.9
, the over-reservation
allows a reduction in the number of signaling events beyond 90 % of that of the
per-flow approach, depending on the network congestion level throughout the
network. Therefore, resource over-reservation techniques are necessary to ensure
