AJI NUR SANYOTO et al: PERFORMANCE EVALUATION OF ROUND ROBIN AND PROPORTIONAL FAIR . .

Performance Evaluation of Round Robin and Proportional Fair Scheduling

Algorithms on 5G Milimeter Wave for Node Density Scenarios

Aji Nur Sanyoto, Doan Perdana, Yoseph Gustommy Bisono

Faculty of Electrical Engineering

Telkom University
Bandung, Indonesia
Email: [email protected], [email protected], [email protected]

Abstract – Numerous imminent challenges face 5G technology, mm-Wave is a potential alternative technology for future networks.
This happens because it offers greater bandwidth, multi-gigabit and low latency wireless links. With these advantages, a MAC
scheduler is needed at the base station to allocate radio resources available for applications in high-speed and real-time data. The
right scheduling algorithm greatly affects network performance. In our research reported in this paper, we compare round robin
and proportional fair algorithm on NS3.27. We determine how the choice of the scheduler has a significant impact on network
performance such as throughput, delay and fairness index. Our results show that round robin is better than proportinal fair: it has
better performance than proportional fair in terms of: a 3.65 % throughput, 18.29% delay and 0.1 % fairness index. We conclude
that round robin is a better choice scheduler for voice traffic.

Keywords - Milimeter Wave, MAC layer, Scheduling Algorithm, Round robin, Proportional fair.

I. INTRODUCTION characteristics of various networks efficiently. The concept

of flexible TDMA is a solution, considering 5G has various
The growth of internet users is increasing rapidly and types of services with very diverse traffic, ranging from
indirectly will lead to an increase of channel capacity. Based applications, devices and usage.
on that, a higher frequency spectrum is required to
accommodate data usage. The Millimeter wave (mm-Wave)
is a wave that work at a frequency range of 30-300 GHz. It is
making Millimeter wave as the central technology in 5G
system because of its potential to achieved massive
throughput required by the future. Nowadays, mm-Wave has
become a key focus of the 3rd Generation Partnership
Project (3GPP) NR1 effort and currently under development
[1]. In accordance with its characteristics, research on mm-
Wave generally uses the 28 GHz band, 38 GHz band, 60
GHz band, and E-band (71-76 GHz and 81-86 GHz) [2]. One
of the requirements for METIS 2020, the target of E2E
latency should be below 10ms, cannot be reach if still using
the previous technology [3]. Therefore, to design of End-to-
End (E2E) cellular systems that can fully utilizes the high-
throughput, low-latency capabilities of mm-Wave network,
innovation will be needed on several layers, such as the
MAC layer. In [4], to support highly directional
transmission, high latency, and high peak level the MAC
layer was redesigned with several developments.
In the current LTE system, the transmission process sent
at a fixed Transmission Time Interval (TTI) of 1ms. On
TDMA scheduling, data allocation becomes inefficient for
small packet.
Therefore in the study [5], a variable TTI-based TDMA
structure has been proposed, also known as flexible TTI.
This scheme allows for varying slot sizes and very suitable
for diverse traffic. In addition, the TTI variable system also
has flexibility in scheduling resources, which can handle the Figure. 1. Flow chart of proportional fair algorithm [7]

DOI 10.5013/IJSSST.a.20.02.17 17.1 ISSN: 1473-804x online, 1473-8031 print

 AJI NUR SANYOTO et al: PERFORMANCE EVALUATION OF ROUND ROBIN AND PROPORTIONAL FAIR . .

The scheduler has a role in the allocation of resource

blocks in transmission between the users with different types
[6]. The main purpose of the packet scheduler algorithm is to
maximize throughput and fairness index.
Figure 1 shows the sequence of how the proportional fair
algorithm works. Channel Quality Indicatior (CQI) values
are mapped according to the type. Then the proportional fair
algorithm calculates based on the value of the average data
rate and throughput in the previous matrix calculation. Then
it is selected with the user who has the highest CQI value. A
schedule is performed on the first user until the slot was full
and continued until the next slot is available.
When managing the user priority on a resource block,
this algorithm stated on m matrix and for user i on a
resource block k as [8]:
d (f)
i, k
m  (1)
i, k R (f)
i

With is average throughput of user i computed in

sub frame f, and is Achievable throughput user k in
m resource block and f sub frame which is a Shannon
expression for the channel capacity as:

d
i, k

( f )  log 1  SNR
i, k
f    (2)

The user’s matrix value above will be compared by

another user’s matrix. User that has a bigger value will be
served first.
Figure 2 shows the sequence of how the round robin Figure. 2. Flow chart of Round Robin algorithm [7]
algorithm works. This algorithm works by rotating the queue
process. Each process has the same time allotment that is In this experiment, we compare scheduler algorithm
equal to time quantum (q). If this quantum time runs out, the between Round Robin (RR) and Proportional Fair (PF) with
server will handle the next process. variable TTI multiuser cell. The purpose of this research was
When managing the user priority on a resource block, to analyze the effect of scheduler on network performance
this algorithm stated on metric m and for user i on a resource such as delay, throughput, and fairness index.
block k as [8]: Furthermore, network performance using the results
m  w t T
i, k i
 i (3)
generated by Network Simulator 3 version 3.27 with the
mm-Wave module developed by [9]. The goal of this
Where notation on the metric (3) known as: research is to analyze the QoS parameters such as
throughput, delay, and Fairness index.
= priority value for every service for user i
t = current time II. RELATED WORK
= last time when user i was served.
Scheduling on cellular technology has been applied to the
The user’s metric value above will be compared by LTE system. Some of them propose and compare between
another user’s metric. User that has a bigger value will be scheduler algorithms. In [10], Mohnish Jha compared the
served first. If the network only has one same service then performance of three schedulers such as Round Robin,
the user will be served like First Come First Served User Priority Set Scheduler and Proportional Fair scheduler by
when the user that comes first will be prioritized. Metric transmitting real-time voice packets with a various channel
from this algorithm will be : conditions of the UE for observing the performance of each

m
i, k
 
 t T
i
(4)
scheduler using NS3.24. The simulation results show that the
round robin scheduler is better compared to the other two in
terms of throughput, delay and jitter at uplink as well as
downlink.

DOI 10.5013/IJSSST.a.20.02.17 17.2 ISSN: 1473-804x online, 1473-8031 print

 AJI NUR SANYOTO et al: PERFORMANCE EVALUATION OF ROUND ROBIN AND PROPORTIONAL FAIR . .

Research about 5G mm-Wave can be done to evaluate

cross-layer and end-to-end performance. In [9], a research
has been conducted on the implementation and validation of
the mm-Wave module in NS-3. They redesigned several
layers because mm-Wave will require innovation not only in
the physical layer, but also across all layers of the
communication protocol stack to fully utilize high
throughput, low latency capabilities and maximum
performance.
In [5] a TTI-based design analysis was conducted and
focused on flexible TTI-based designs, in terms of how well
they utilized the allocated radio resources, and found that
flexible frame structures exceeded fixed structures in all Figure. 3. Simulation Topology
traffic scenarios discussed, especially for small burst traffic.
So it can be concluded that the flexible TTI scheme will be The system design is shown in Figure 4. The remote host
very suitable to be applied on mm-Wave communication. node has a function as the sender and connected to the PGW
R. Ford analyze the benefits of variable TTI over fixed node in point to point mode. Datarate between PGW and
TTI on [3]. The simulation is done using ns-3 full-stack Remote host is 100Gbps with 0,01s delay. PGW was
simulation model for mm-Wave cellular networks with 1 connected to eNodeB using the LTE EPC core network,
GHz bandwidth. They evaluate latency performance and which indicates that the network to be simulated is a non-
found that flexible TTI has lower latency value than fixed standalone 5G network.
TTI. This happens because on the fixed TTI, all subframes The parameters and its description are shown in Table 1.
are allocated to one user, but for the flexible TTI, schedulers The scenario in this research is to change node density
can allocate a data symbols in a subframe that matches for simulated by different scheduling algorithms.
each user. Changes in the number of nodes in the scenario vary
Several studies about scheduling on 5G networks also from twenty to one hundred with intervals of 20 UEs. The
have been researched before. K. Gomez on research [11], scheduler algorithm that will be used is round robin and
provided a comparative study of a different scheduling proportional fair. UEs positions are arranged randomly with
disciplines that can be used in future 5G especially on constant position mobility model.
emergency communications for public safety. In addition to TABLE I. SIMULATION PARAMETERS
proposing a new disciplinary scheduler, simulation results Parameter Quantity
show that the proportional fair scheduler can be the Mmwave Carrier frequency 28 GHz
beneficial for both cell throughput and fairness when serving Mmwave Bandwidth 1 GHz
an emergency communication and commercial users at the Number of eNodeB 1
Number of User 20, 40, 60, 80, 100
same time. User Mobilility Constant Position
In this research, two scheduler algorithms that have been Datarate 8 Kbps
adjusted to the flexible TTI scheme in the NS3 mm-Wave Packet Size 20 bytes
module will be compared. The parameters to be analyzed are Transport Layer UDP
network performance in terms of throughput, delay, and Scheduler Algorithm Round robin, Proportional Fair
fairness index.
The number of packet size and data rate used in the
III. RESEARCH METHOD simulation is adjusted to the characteristics of the packet size
and data rate on one of the VoIP codecs G.729
The simulations on this research were performed on the Voice traffic is simulated by adding OnOffHelper. The
Network Simulator 3.27 with an additional mm-Wave constant packet is transmitted periodically as long as the time
module. The mm-Wave module is designed for end-to-end is ON when the user speaks. Packages are not sent during
simulations of 3GPP style cellular networks. In general, this OFF time and the user stops talking. The UDP Transport
research will discuss about the effect of scheduler algorithms Protocol was chosen because it is in accordance with the
on network performance with density changes using voip need for data transmission in VoIP communications where
traffic. data delivery time is important [12].

DOI 10.5013/IJSSST.a.20.02.17 17.3 ISSN: 1473-804x online, 1473-8031 print

 AJI NUR SANYOTO et al: PERFORMANCE EVALUATION OF ROUND ROBIN AND PROPORTIONAL FAIR . .

Where notation on the formula (6) known as:

Trx = Time of received packet on destination

Ttx = Time of packet send on source
= Received packet

Fairness Index, defined as the level of fairness of

scheduling algorithms in schedule packages and allocation
of resources to be sent. The theory and formula regarding
the fairness index was revealed by [14]. Metrics of the
formula are known as Jain's Fairness Index. Maximum value
of this metrics is 1, where it indicates perfect fairness among
the users in the system.
2
 n 
  x
 i
f ( x) 
i  1  (7)
n
2
n  x
i
i 1
Where notation on the formula (7) known as:

f(x) = fairness index

n = number of user
x = Throughput user i

IV. RESULT AND ANALYSIS

After simulating voice and video traffic from the 5G mm-

Wave network in NS3, we obtained performance results such
as throughput, delay and fairness index, then be analyzed.
Figure. 4. Flowchart system The analysis divided into two parts for voice and video
traffic to find out which better scheduler for the two services.
Figure 4 shows the flowchart system. After designing the
module in the NS3 environment, the simulation design is A. Delay Evaluation
adjusted to the scenario. The scheduling algorithm is
implemented and simulated alternately. Changes on the
number of nodes are set gradually from 20, 40, 60, 80, and
100 nodes. If the simulation is failed, the simulation scenario
design will be reconfigured. If it is successful, the output that
will be analyzed are throughput, delay, and fairness index.
Throughput, defined as the effective ability of a network
in sending data. Throughput is the number of packets
received in bits divided by the amount of delivery time [13].

Throughput 
 Rx Packet Size (5)
Delivery Time

Delay, defined as the time it takes for a package to Figure. 5. Delay on change number of users for voice traffic
deliver from source to the destination. The delay value starts
calculated when the source starts sending packets and ends Figure 5 shows the effect when increasing the number of
when the destination actually receives the packet . users to the delay obtained from the voice traffic simulation.
The lowest delay in round robin occurred on 20 UE with
Trx  Ttx
Delay  (6) 1.023ms, for proportional fair lowest delay occurred on 20
 Rx

DOI 10.5013/IJSSST.a.20.02.17 17.4 ISSN: 1473-804x online, 1473-8031 print

 AJI NUR SANYOTO et al: PERFORMANCE EVALUATION OF ROUND ROBIN AND PROPORTIONAL FAIR . .

UE with 1.285ms. On 100 UE, round robin and proportional simulation of adding the number of users to the proportional
fair generating the highest delay with 1.321ms and 1.755ms. fair algorithm is 0.994. It is lower than the round robin’s
Average delay obtained from round robin is 1.215ms. fairness index that has 0.995. The value obtained by the
This is 18.29 % lower than proportional fair with average round robin algorithm is greater because this algorithm does
delay of 1.487ms. Based on figure 5, it can be conclude that not consider channel conditions so that it offers a higher
delay for both scheduler increase, due to increase of number fairness value.
of UE make waiting time for each users to be served is
getting longer.
Round Robin has a better delay because for small
packages, users queuing don’t to take long time, different
with Proportional fair which must take consider the channel
quality.
Proportional fair has higher delay because this algorithm
is considering the Channel Quality Indicator (CQI) value
when deciding its algorithm matrix. In this case, UE has a
different channel condition, and on the other side this
algorithm has to serve all UE while maintaining its fairness
so the delivery took a longer time.
Figure 7. Fairness Index on change number of users for voice traffic
B. Throughput Evaluation
Round robin and proportional fair have a decreasing
fairness index value against the increase in the number of
users. This happened because of the increasing number of
users, more users were served which reduced the value of
fairness. Both schedulers show fairness due to the fairness
index being close to 1.

V. CONCLUSION

In this paper, our work focused on scheduling in a 5G

mm-Wave network with a new MAC layer structure that was
proposed in previous studies. Based on simulation results,
the choice of scheduling algorithm has an effect on network
Figure 6. Throughput on change number of users for voice traffic performance. Evaluating the use of a scheduler, on a flexible
TTI scheme applied to a 5G mm-Wave network using VoIP
Figure 6 shows effect the when increasing the number of traffic, shows that the round robin algorithm is better in
users to the throughput obtained from the voice simulation. It terms of throughput, delay and fairness index than the
show that the increase number of users, throughput proportional fair algorithm. For the delay, Round Robin
decreased due to the bandwidth capacity will be shared with algorithm obtained 1.215ms on average. This is 18.29%
all users. lower than proportional fair average delay with 1.487ms. For
Round robin gets average throughput of 0.137 Mbps. throughput, Round Robin algorithm obtained 0.137 Mbps on
This is 3.65 % better than proportional fair with average average. It is 3.65% better than the proportional fair’s
throughput of 0.132 Mbps. The lowest throughput in throughput which has 0.132 Mbps. For fairness index
proportional fair occurred on 100 UE with 0.131 Mbps, and Proportional fair algorithm obtained 0.994 on average, and
for round robin occurred on 100 UE with 0.132 Mbps. round robin has 0.995. Both schedulers show fairness due to
Round robin has a higher throughput because this the fairness index close to 1.
algorithm not consider the channel condition and has main
purpose to balance between throughput and fairness among REFERENCES
all the users. The proportional fair’s delay has higher delay,
that makes the packet received is less than round robin. [1] “NR and NG-RAN overall description—Rel. 15,” 3GPP, Sophia
Antipolis, France, Rep. TS 38.300, 2018.
C. Fairness Index Evaluation [2] Niu, Y., Li, Y., Jin, D., Su, L., & Vasilakos, A.V. “A Survey of
Millimeter Wave (mmWave) Communications for 5G: Opportunities
and Challenges” arXiv:1502.07228 [cs.NI], 2015.
Figure 7 shows the effect when increasing the number of [3] R. Ford, M. Zhang, M. Mezzavilla, S. Dutta, S. Rangan and M.
users to fairness index in each scheduling algorithm for voice Zorzi, "Achieving Ultra-Low Latency in 5G Millimeter Wave
simulation. The average of fairness index obtained from the Cellular Networks," in IEEE Communications Magazine, vol. 55, no.
3, pp. 196-203, March 2017.

DOI 10.5013/IJSSST.a.20.02.17 17.5 ISSN: 1473-804x online, 1473-8031 print

 AJI NUR SANYOTO et al: PERFORMANCE EVALUATION OF ROUND ROBIN AND PROPORTIONAL FAIR . .

[4] S. Dutta, M. Mezzavilla, R. Ford, M. Zhang, S. Rangan and M. Simulation of 5G mmWave Networks", Communications Surveys &
Zorzi, "MAC layer frame design for millimeter wave cellular Tutorials IEEE, vol. 20, no. 3, pp. 2237-2263, 2018.
system," 2016 European Conference on Networks and [10] Mohnish Jha et al, “Comparative Analysis of MAC Scheduling
Communications (EuCNC), Athens, pp. 117-121, 2016. Algorithms in Long Term Evolution Networks using NS3”. Asian
[5] Sourjya Dutta, Marco Mezzavilla, Russell Ford, Menglei Zhang, Journal of Enginnering Technology and Innovation, Vol 4(7):124-
Sundeep Rangan, Michele Zorzi, "Frame Structure Design and 127, 2016.
Analysis for Millimeter Wave Cellular Systems", Wireless [11] K. Gomez, L. Goratti, F. Granelli and T. Rasheed, "A comparative
Communications IEEE Transactions on, vol. 16, no. 3, pp. 1508- study of scheduling disciplines in 5G systems for emergency
1522, 2017. communications," 1st International Conference on 5G for Ubiquitous
[6] de Oliveira, R.P. & de Góis, L.A. & Foronda, A., “Enhanced PF Connectivity, Akaslompolo, pp. 40-45, 2014.
scheduling algorithm for LTE networks”, International Journal of [12] C. Olariu, “Quality of Service Support for Voice over IP in Wireless
Communication Networks and Information Security (IJCNIS), vol. Access Networks,” Waterford Institute of Technology, 2013.
10, no. 1, 2018. [13] T. Wulandari, D. Perdana and R. M. Negara, "Node Density
[7] Mannani, Dinesh, “Modeling and Simulation of Scheduling Performance Analysis on IEEE 802.11ah Standard for VoIP
Algorithms in LTE Networks”, Warsawa, Warsaw University of Service," International Journal of Communication Networks and
Technology, 2012. Information Security (IJCNIS), vol. 10, no. 1, 2018.
[8] F. Capozzi, G. Piro, L. A. Grieco, G. Boggia and P. Camarda, [14] M. Carpin, A. Zanella, J. Rasool, K. Mahmood, O. Grøndalen and
"Downlink Packet Scheduling in LTE Cellular Networks: Key O. N. Østerbø, "A performance comparison of LTE downlink
Design Issues and a Survey," in IEEE Communications Surveys & scheduling algorithms in time and frequency domains," 2015 IEEE
Tutorials, vol. 15, no. 2, pp. 678-700, 2013. International Conference on Communications (ICC), London, pp.
[9] Marco Mezzavilla, Menglei Zhang, Michele Polese, Russell Ford, 3173-3179, 2015.
Sourjya Dutta, Sundeep Rangan, Michele Zorzi, "End-to-End

DOI 10.5013/IJSSST.a.20.02.17 17.6 ISSN: 1473-804x online, 1473-8031 print