Page 224 - ITU Kaleidoscope 2016

2016 ITU Kaleidoscope Academic Conference

The main conclusion, comparing with the results
obtained with the emulated testbed or simulation
environment, is that the impact is even harder for the strong
implication of Hybrid Slow-Start’s poor performance with
delay variability. The more realistic the delay over the
system, the more challenging conditions of mobile network
and therefore, more complex to adapt to. Different steps in
the experimentation stair confirm the detected performance
of Hybrid Slow-Start but denote a very distinct impact.
Secondly, related to longer TCP transmissions in which
congestion avoidance phase has a crucial role to play in the
adaptability to the available capacity, different CCAs
(a) propose distinct ways to increase and decrease the
congestion window (CWND) in order to achieve maximum
throughput. In this regard, it is not straightforward to
propose QoS measurements that are able to consider the
different procedural way each CCA has.
As explained in the Section 1, many are the factors that
can affect the available capacity. In ns-3 we studied the
impact that the mobility inertia (positions from and to) and
speed could have on CCA’s ability to take advantage of all
radio resources. We deployed a scenario in which the UE
moved from the eNodeB to further positions and the other
way around. During the transmission we studied for each
MCS range the number of times that each CCA wasted

transmission opportunities due to the impossibility to
(b) constantly feed the network (Figure 7). The same tests were
Figure 5. Slow-Start algorithms comparison (emulated testbed): performed with different speeds (60 kilometers per hour -
a) CWND evolution; b) Impact on throughput kmph- and 200 kmph). Figure 7 aims to show differences
among CCAs under different speeds at a glance without
The last comparison and verification example of the
addressing very precise effects. For that reason, the two
impact of Hybrid Slow-Start was carried out in a LTE
movement inertias are not necessary and only the backward
controlled deployment. To the correct dissemination of the
movement is depicted. The work [26] concluded that indeed
results, 3 tests were launched in each mobility pattern. In
the speed had an impact, the need for studying similar
this case the representation of final throughput is illustrative
effects in real deployments was highlighted due to the
enough to understand the effect. necessity to measure all the involved effects such as

propagation delay, processing, queuing, variable fading
model and delay and so forth.

Figure 6. Slow-Start algorithms comparison (controlled
deployment): Impact on throughput

The outcome of Figure 6 represents a huge impact of Figure 7. Different CCAs’ performance at different speeds [26].

Hybrid Slow-Start on final achieved throughput regardless
Even though all micro-effect were not considered, some
the selected mobility pattern. The whole deployment
important and macroscopic features were addressed
described in Section 2.1.3 is composed of real equipment
regarding CCAs behavior over LTE. Firstly, Westwood+
and the queuing and processing delay and interaction
showed a painful deficiency in cellular networks regarding
constraints are the closest to real-world deployments ones.
its incapability to estimate the available bandwidth properly
while applying the back-off, causing a huge constraint to be

– 206 –