Page 222 - ITU Kaleidoscope 2016

2016 ITU Kaleidoscope Academic Conference

[23] to define TCP-based measurements’ constraints and (Received signal strength indication -RSSI- , Reference
ways to overcome them. The role of WS negotiation as a Signal Received Power –RSRP- , Reference Signal
static constraint that may establish a ceiling in the maximum Received Quality -RSRQ- or Signal to noise ratio –SNR-)
achievable effective window size (and therefore goodput) provide accurate mappings to CQI, neither for mobility nor
was shown. In some cases slow-start phase may be enough static patterns, resulting in up to 10 units of CQI variability
to fill large enough bottleneck link queue, resulting in stable for a single value. 2) Even if the CQI value could be
goodput from that point on. However, in high bandwidth- inferred precisely from physical measurements, several
delay-product (BDP) cases, it is no longer true. Therefore, effects in commercial implementations prevent a direct
the necessity to consider congestion epoch times as a stop mapping to upper layer performance. For instance,
criteria to ensure the achievement of the maximum available commercial UEs usually implement techniques such as
capacity (and accuracy of the estimation) has been Outer Loop Link Adaptation (OLLA) in order to gain or
identified. In order to shorten the test duration, the use of ensure robustness in its communication. In the same way,
multi-thread was suggested based on the different tests and under low traffic demands, commercial eNodeB’s, taking
measurements carried out. into account reported CQI, usually assign lower level of
All the aforementioned advances propose QoS modulation and coding scheme (MCS) for robustness
measurements that are highly focused on fixed networks purposes.
features. Nowadays, QoS measurements should be adapted These findings address the need for real UEs study so
to mobile networks and to achieve that goal the as to involve all aforementioned constraints, features and
understanding of TCP under different cellular network inaccuracies in the process and understand the effects that
circumstances is essential. Our work has involved end-users may well suffer. Considering the emulated testbed
simulation environment, emulation testbed and controlled as the first step in the experimentation stair that allows
deployment and it has been devoted to this end with a triple performing with real UEs, the usage of such testbed for a
purpose: a) Study TCP over LTE so as to provide real- proper analysis is likely to be minimal.
world deployments with the knowledge to overcome Concerning the substantial importance of UE’s and
constraints. b) Give insights to understand cross-layer eNodeB’s, we have measured the reported quality in many
behavior and be able to reason either TCP-based or model- different conditions divided in two testing frameworks. In
based outcome. c) Detect extended deficiencies and verify the emulated testbed we have modeled the channel
them in different experimentation stair’ layers (Figure 1). conditions of two UEs (dongle and mobile phone) from
their reported CQI in six of the LTE testing points that the
3. APPLICABILITY TO PERFORMANCE TESTING 3GPP suggests. Table 2 gathers such testing points and their
OVER MOBILE NETWORKS network conditions (considering different situation amongst

This section covers the applicability analysis of the the fading profiles of Extended Pedestrian A model –EPA-,
described experimentation frameworks and methodologies Extended Vehicular A model –EVA- and Extended Typical
for carrying out performance tests in the specific field of Urban model (ETU)). All testing-points are static and
TCP over mobile broadband networks. Specifically, the adequate for emulated environments to avoid mobility
general challenges are: constraint in such testbeds (as stated in section 2.1.2).
• The impact of using real-world equipment (clients, Although there is a possibility of emulating UE’s
servers and network nodes) for performance tests. movement through the introduction of variable SINR traces,
it is not a real movement and neither the fading and nor the
• The impact of including different models to
propagation represent a purely realistic behavior so as to
characterize the variable behavior of delay and bitrate
in short and long term. achieve a real variable bitrate and delay. This issue could be
solved by comparing with realistic performance and finding
• The impact in the different phases of TCP operation.
In this regard, we analyze the importance on using real common patterns, but cannot be studied in an isolated way.

UEs, the impact of different devices, the relevance of real Table 2. LTE testing points (3GPP)
fading and propagation and the impact of small-scale cells. No. Propagation SNR
Following the different steps of the experimentation stair, Condition
some TCP features are analyzed in order to make a 1 EPA5 20dB
2 EVA5 10dB
comparative study. Even though the study is purely TCP- 3 EVA70 20dB
based, the understanding of such behaviors would be able to 4 EVA200 20dB
provide model-based measurements with more accurate 5 ETU70 10dB *
input parameters. 6 ETU300 10dB *
*Original testing point were placed in 0dB (edge of the cell).
Adaptation to the middle of the cell with 10dB.
3.1. Impact of real-world equipment
Having in mind the abovementioned testing points,
According to the analysis in [24], in which several Figure 2 shows as empirical cumulative distribution
channel quality measurement methodologies are evaluated,
function (ECDF) the differences that the reporting of
none of the studies provided adequate channel quality
selected UEs have between each other in exactly same
information to the timescales required by a LTE scheduler.
conditions with sufficient repetitions (5) and logging time
Two are the major issues: 1) None of the radio parameters
(120 seconds). It is clear that for exactly the same network

– 204 –