저 시-비 리- 경 지 2.0 한민

는 아래 조건 르는 경 에 한하여 게

l 저 물 복제, 포, 전송, 전시, 공연 송할 수 습니다.

다 과 같 조건 라야 합니다:

l 하는, 저 물 나 포 경 , 저 물에 적 된 허락조건 명확하게 나타내어야 합니다.

l 저 터 허가를 면 러한 조건들 적 되지 않습니다.

저 에 른 리는 내 에 하여 향 지 않습니다.

것 허락규약(Legal Code) 해하 쉽게 약한 것 니다.

Disclaimer

저 시. 하는 원저 를 시하여야 합니다.

비 리. 하는 저 물 리 목적 할 수 없습니다.

경 지. 하는 저 물 개 , 형 또는 가공할 수 없습니다.

M.S. THESIS

Split Algorithm with a PDCP ReorderingFunction for LTE/mmWave Dual

Connectivity

LTE/mmWave이중연결성을위한분할알고리즘과PDCP재배열기작

BY

DONGYEON WOO

February 2018

DEPARTMENT OF ELECTRICAL ENGINEERING ANDCOMPUTER SCIENCE

COLLEGE OF ENGINEERINGSEOUL NATIONAL UNIVERSITY

M.S. THESIS

Split Algorithm with a PDCP ReorderingFunction for LTE/mmWave Dual

Connectivity

LTE/mmWave이중연결성을위한분할알고리즘과PDCP재배열기작

BY

DONGYEON WOO

February 2018

DEPARTMENT OF ELECTRICAL ENGINEERING ANDCOMPUTER SCIENCE

COLLEGE OF ENGINEERINGSEOUL NATIONAL UNIVERSITY

Abstract

On the transition toward 5G communication service from LTE communication

service, Non-Stand Alone scenario is required. Generally, when two different base

stations are available, it seems like better throughput and reliability is expected, but it

needs well-coordinated controlling algorithm.

In this paper, we consider LTE/mmWave dual connectivity to find a way to make

use of two different radio technology, LTE and mmWave. To utilize both links simul-

taneously, we propose a split algorithm that includes a traffic split operation and a

packet reordering operation. And, the proposed split algorithm is evaluated through

ns-3 simulation.

keywords: dual connectivity, LTE, mmWave, multi-path transmission

student number: 2016-20936

i

Contents

Abstract i

Contents ii

List of Figures iv

1 INTRODUCTION 1

2 LTE/mmWave Dual Connectivity 3

2.1 Dual Connectivity Architecture . . . . . . . . . . . . . . . . . . . . . 3

2.2 PDCP Reordering Function . . . . . . . . . . . . . . . . . . . . . . . 4

3 Split Algorithm Design 6

3.1 Split Ratio Formulation . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.2 Split Ratio Implementation . . . . . . . . . . . . . . . . . . . . . . . 7

3.3 EarlyDropTimer Design . . . . . . . . . . . . . . . . . . . . . . . . 8

4 Performance Evaluation 9

4.1 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4.2 Evaluation of Split Ratio . . . . . . . . . . . . . . . . . . . . . . . . 9

4.3 Evaluation of EarlyDropTimer . . . . . . . . . . . . . . . . . . . . . 11

4.4 Evaluation of Split Algorithm Design . . . . . . . . . . . . . . . . . 13

5 Conclusion 15

ii

Abstract (In Korean) 17

iii

List of Figures

2.1 LTE/mmWave Dual Connectivity Architecture. . . . . . . . . . . . . 4

3.1 Traffic Splitting with a Split Ratio η. . . . . . . . . . . . . . . . . . . 7

4.1 Split Ratio and Throughput Transition when UDP Source Rate Changes. 10

4.2 Average Path Delay for Different UDP Source Rate. . . . . . . . . . . 10

4.3 TCP 2000 Mb/s without EarlyDropTimer. . . . . . . . . . . . . . . . 12

4.4 TCP 2000 Mb/s with EarlyDropTimer. . . . . . . . . . . . . . . . . . 12

4.5 TCP transmission completion time. . . . . . . . . . . . . . . . . . . . 13

4.6 Web Page Load Time. . . . . . . . . . . . . . . . . . . . . . . . . . . 14

iv

Chapter 1

INTRODUCTION

As technologies such as virtual reality or augmented reality are being interested, and as

smart machines are widely deployed, the need for new communication services such

as enhanced Mobile Broadband (eMBB), massive MTC (mMTC) or Ultra-Reliable

Low Latency Connectivity (URLLC) is appearing [1]. Currently, standardization of

5G communication service is being discussed with several innovations including 5G

New Radio (NR).

It is considered to eventually replace current LTE with 5G NR, but during the

transition toward 5G NR, it is required to support Non-Stand Alone (NSA) operation

to ensure practicality. This issue is under big consideration, and NSA NR is decided

as one of the target contents for Release-15 work item [2].

Following current issues, our research is focusing on NSR NR scenario, especially

LTE/mmWave dual connectivity (DC). mmWave is a radio access technology utiliz-

ing above 6 GHz spectrum bands. In LTE/mmWave DC, user equipment (UE) estab-

lishes connections with both LTE and mmWave base stations, concurrently. In this

architecture, our key question is about how to achieve performance improvement on

LTE/mmWave DC.

Typically, when multi-path connections are available in a network, it looks ben-

eficial to activate both connections concurrently. However, under the condition with

1

notable link capacity difference, the situation is different. For LTE and mmWave DC,

peak data rate is hundreds of megabits per second for LTE and several gigabits per sec-

ond for mmWave. With this high capacity difference, it is hard to expect throughput

gain by using two paths simultaneously, according to the difficulty of traffic controlling

with heterogeneous paths. Also, there is another factor which needs to be considered,

X2 interface delay, the packet transmission delay between two base stations.

To derive benefit from LTE/mmWave DC, we propose a split algorithm which

includes a split ratio formation and a PDCP reordering function. The split ratio is

calculated based on the service rate estimation to forward traffic to one of base stations

in right amount, and the revised PDCP reordering function includes the concept of

packet early dropping which is inspired by CoDel routing scheme [3]. In following

chapters, we will show detailed description of the proposed split algorithm with the

evaluation results.

2

Chapter 2

LTE/mmWave Dual Connectivity

2.1 Dual Connectivity Architecture

Through this research, we consider a LTE/mmWave dual connectivity (DC) with 3C

architecture [4]. In LTE/mmWave dual connectivity, a user equipment (UE) makes

connections with an LTE base station and a mmWave base station. On the enhanced

packet core (EPC) side, connections are anchored by the LTE base station considering

3C architecture, and we denote the LTE base station as master eNB (MeNB) and the

mmWave base station as secondary eNB (SeNB). The detailed architecture is shown

in Figure 2.1

For 3C DC architecture, three different kinds of bearers can be activated, a bearer

passing only MeNB, passing only SeNB, or passing both MeNB and SeNB. The bearer

constructed through both MeNB and SeNB is called split bearer, and since the split

bearer activates two distinct paths simultaneously, there are two characterized opera-

tions required. On the anchoring point of the sender side, the traffic split is required

to decide a path to route packets. On the gathering point of the receiver side, packet

reordering is required to guarantee in-order delivery of packets.

On the following analysis, we use both UDP and TCP traffics for simulation. Note

that UDP and TCP traffics have different requirement at the receiver, TCP needs to

3

Figure 2.1: LTE/mmWave Dual Connectivity Architecture.

guarantee in-order delivery, while UDP does not need to do so. So, we will use UDP

traffic to evaluate only the split ratio formulation, and use TCP traffic to evaluate the

aggregated performance of split ratio formulation and reordering function.

To connect MeNB and SeNB, X2 interface locates between them. When traffic

from MeNB is forwarded toward SeNB, the traffic pass through the X2 interface. Since

this X2 interface also causes additional propagation delay, the X2 interface delay needs

to be considered.

2.2 PDCP Reordering Function

3GPP already specified the PDCP reordering function to support LTE dual connectiv-

ity or LTE-WLAN aggregation (LWA) [5]. The basic idea of this function is to main-

tain a buffer in receiver’s PDCP layer to store incoming packets for a while and then

deliver packets to the upper layer in sequence. Since all packets have serial number

4

attached at MeNB, the receiver’s PDCP buffer is able to arrange them with in-order

sequence. So, packets in serial sequence are confirmed to be delivered to upper layer,

and disarranged packets should stay in the buffer.

When a packet stays in the reordering buffer for too long time, PDCP Reorder-

ing Timer expires. Then, disarranged packets are delivered and there is an issue that

lots of packet loss are detected in the upper layer which leads to malfunction of TCP

congestion control.

5

Chapter 3

Split Algorithm Design

As mentioned in the previous chapter, to construct and utilize a split bearer, two core

operations are required in the MeNB side and the UE side. To maximize the utilization

of the split bearer, we propose a split algorithm which includes both a traffic split

operation and a packet reordering operation. The detailed algorithm will be shown

below.

3.1 Split Ratio Formulation

For the purpose of routing incoming packets toward MeNB or SeNB, we introduce

a term split ratio η to denote the ratio of the traffic routed toward MeNB to the total

traffic. By deriving the adequate split ratio value, it is possible to maximize utilization

of dual paths.

To decide the split ratio value, our intuition is to ‘send packets to a path with

shorter path delay.’ Since the main cause of packet reordering delay is delay difference

between two paths, by sending packets to faster path, it is possible to minimize the

packet reordering delay.

Then, the specific procedure of deciding split ratio value can be represented as

follows,

6

Figure 3.1: Traffic Splitting with a Split Ratio η.

• Find η minimizing max(Qt+1

M

µt+1M

,Qt+1

S

µt+1S

+ TX2

)when η denotes split ratio, µtM , µ

tS denotes RLC service rate of MeNB and

SeNB, QtM , QtS denotes RLC queue size of MeNB and SeNB, TX2 denotes X2

interface delay, and N denotes the number of incoming packets

• Since the queue size is a function of the previous queue size, the number of

incoming packets and the number of serviced packet, we know that

QtM = Qt−1M + ηtN t − µtM (3.1)

QtS = Qt−1S +

(1− ηt

)N t − µtS (3.2)

• Then, we can derive a closed form of the split ratio value

ηt+1 =µtM

(QtS +N t+1

)− µtSQ

tM + µtMµ

tSTX2

N t+1(µtS + µtM

) (3.3)

3.2 Split Ratio Implementation

Deciding routing path of every individual incoming packet is the ideal method, but it

requires too much load in collecting queue information and computing exact value.

So, we suggest calculating the split ratio in every fixed interval.

The specific procedure is as follows,

7

1. For every SplitTimerInterval, MeNB RLC layer and SeNB RLC layer sends

their transmission queue sizes to MeNB PDCP layer where the traffic split oc-

curs. Typically 10ms of SplitTimerInterval is adopted in our design considering

scheduling time unit at each eNB.

2. In MeNB PDCP layer, split ratio value is calculated using the closed form equa-

tion.

3. Incoming packets are forwarded according to the split ratio value until the value

is updated. To divide the finite number of packets according to the split ratio, we

group a number of packets into a single chunk, and divide the chunk into two

different paths.

3.3 EarlyDropTimer Design

As we mentioned before, PDCP reordering function has its own PDCP Reordering

Timer to handle packet losses, and when the timer expires, it is observed that bunch of

packet losses is occurred simultaneously. Although the reordering timer rarely expires,

when it occurs, TCP detects the losses and the congestion window goes to slow start

phase. And during the slow start phase, the TCP congestion window needs to grow

from the bottom degrading the whole TCP throughput dramatically.

To resolve this issue, we revise the PDCP reordering function by adding an addi-

tional timer, EarlyDropTimer. The operation of EarlyDropTimer is quite simple. The

EarlyDropTimer is set in the same way with PDCP Reordering Timer, but has shorter

expiration time. When the EarlyDropTimer expires, it drops a single packet intention-

ally. When TCP detects the single packet loss, it does not go to slow start, but just

decrease the size of congestion window to the half. So, it is expected the EarlyDrop-

Timer to control TCP congestion window indirectly before TCP goes to slow start.

8

Chapter 4

Performance Evaluation

4.1 Simulation Setup

We use ns-3.26 network simulator to evaluate our split algorithm design. LTE/mmWave

dual connectivity architecture [6] is implemented using ns-3 lena module and NYU

mmWave module [7]. To implement dual connectivity, additional PHY, MAC, RLC

layers are installed inside UE’s netdevice, and SeNB constructs connection with the

secondary PHY, MAC, RLC layers. Moreover, X2 interface is installed between MeNB

PDCP layer and SeNB RLC layer.

4.2 Evaluation of Split Ratio

As a first step of evaluation, it is necessary to verify how the real traffic is adjusted

when the split ratio changes. Figure 4.1 shows the rough tendency of path through-

put according to the split ratio, when the traffic type is UDP and the source rate is

maintained as 500, 1000 and 200 Mb/s for two seconds each.

It is found that the split ratio follows the source rate by sending more traffic toward

SeNB when higher source rate is being serviced. Following this split ratio, SeNB path

throughput varies a lot while MeNB path throughput stays in a relatively stable rate.

9

Figure 4.1: Split Ratio and Throughput Transition when UDP Source Rate Changes.

Figure 4.2: Average Path Delay for Different UDP Source Rate.

10

Figure 4.2 shows a quantitative analysis of the split ratio. It measures the average

path delay of UDP traffic with source rate 10, 50, 100, 500 and 1000 Mb/s. Since our

split ratio aims to minimize the average path delay, the proposed algorithm shows the

minimum or close to minimum average path delay for all cases.

More specifically, because of the existence of X2 interface delay, for small UDP

traffic, it is more profitable to forward more traffic to MeNB. However, for large UDP

traffic, forwarding more traffic to SeNB is better because MeNB cannot handle hun-

dreds of Mb/s traffic. Figure 4.2 also shows such tendencies.

Our proposed algorithm which is able to adopt split ratio value as the traffic amount

changes shows good performance for all traffic cases. The average path delay of the

proposed scheme almost reaches the minimum value. But, there are small performance

degradation comparing with the best cases, because the proposed algorithm needs to

send at least little amount of traffic to both paths to estimate the service rate for each

path.

4.3 Evaluation of EarlyDropTimer

Figure 4.3 and Figure 4.4 show throughput and TCP congestion window with and with-

out EarlyDropTimer. TCP server is sending traffic with maximum source rate 2000

Mb/s. MeNB path throughput and SeNB path throughput are measured at RLC layer

to distinguish traffic on each path separately. Total throughput which is denoted as

’MeNB+SeNB’ is measured at UE-side PDCP layer, so the effect of reordering oper-

ation is shown as small fluctuations.

Around 2.8 seconds, without EarlyDropTimer, TCP congestion window goes to

slow start and the throughput drops to zero. In the other hand, with EarlyDropTimer,

TCP congestion window does not go to slow start, and maintains the throughput in

relatively stable state. From this result, we can verify that by using EarlyDropTimer, it

is possible to prevent TCP from going into slow start phase.

11

Figure 4.3: TCP 2000 Mb/s without EarlyDropTimer.

Figure 4.4: TCP 2000 Mb/s with EarlyDropTimer.

12

4.4 Evaluation of Split Algorithm Design

To evaluate the overall performance of the proposed algorithm, we measured the time

required to transmit 10 KB, 100 KB, 1 MB, 10 MB, 100 MB and 1 GB size TCP

traffic.

Figure 4.5: TCP transmission completion time.

Figure 4.5 shows similar tendencies with the evaluation result of UDP path delay.

When the traffic size is small, it is more profitable to forward more traffic toward

MeNB, while for larger traffic size, the portion of SeNB increases.

Our proposed algorithm shows almost same performance with split ratio 1 case for

1 MB or smaller traffic, and shows similar performance with split ratio 0 case for 10

MB or larger traffic, which means that adjusting split ratio is effective.

The performance gain obtained from the proposed split algorithm is also observ-

able under actual application. Web page load time is one of the well known metric

used for evaluating the network performance. To simulate the web browsing, we ob-

tained the trace of a web page, www.naver.com, and built a trail of http requests and

responses. The built trace is formed as a series of 16 distinct TCP traffic.

Figure 4.6 shows the measured web page load time. Since http requests and re-

sponses are consisted of small TCP traffic, it is expected for the proposed algorithm to

13

Figure 4.6: Web Page Load Time.

show shorter load time comparing with other comparing cases, and it actually shows

up to 17 percent of performance improvement.

14

Chapter 5

Conclusion

LTE/mmWave dual connectivity is coming closer as 5G standardization progresses,

but not many researchers are taking attention on it. To support the transition toward 5G

service and to achieve better performance we focused on LTE/mmWave DC scenario

and designed a split algorithm which includes a traffic split operation and a revised

packet reordering operation. By ns-3 simulation, we verified each functions and the

integrated module under various traffic conditions and with various metrics. Also, we

succeeded in analyzing the performance of proposed algorithm under different traffic

conditions.

As a next step, different scenarios can be considered. When more mmWave base

stations are available, mmWave/mmWave dual connectivity architecture is possible.

And, depending on the anchoring node, there are also more variations. Under these

various scenarios, it would be possible to supplement the mmWave-only connection

which has high energy cost and high propagation loss by utilizing multi-path selec-

tively.

15

Bibliography

[1] IMT Vision, “Framework and overall objectives of the future development of

IMT for 2020 and beyond“, ITU-R Recommendation M.2083-0, September

2015.

[2] RP-170741, 3GPP RAN 75, March 2017.

[3] Nichols, Kathleen, and Van Jacobson, “Controlling queue delay,“ Communica-

tions of the ACM., 55(7), pp. 42-50. ACM, 2012:

[4] TR 36.842, “Study on Small Cell enhancements for E-UTRA and E-UTRAN;

Higher layer aspects,“ Release 12, 3GPP.

[5] TS 36.323, “Evolved Universal Terrestrial Radio Access (E-UTRA); Packet Data

Convergence Protocol (PDCP) specification,“ Release 14, 3GPP.

[6] LTE/mmWave Dual Connectivity Simulator,

https://github.com/netlab5G/LteMmwaveDc

[7] Mezzavilla, Marco, Sourjya Dutta, Menglei Zhang, Mustafa Riza Akdeniz and

Sundeep Rangan, “5G mmwave module for the ns-3 network simulator,” Pro-

ceedings of the 18th ACM International Conference on Modeling, Analysis and

Simulation of Wireless and Mobile Systems., pp. 283-290. ACM, 2015.

16

초록

LTE 통신 서비스에서 5G 통신 서비스로의 진화 과정에 있어서, 기존의 LTE와

5G New Radio가 공존하는 Non-Stand Alone (NSA) 환경은 필연적으로 고려되어

야 한다. 이와 같이 LTE와 5G New Radio 두 가지 기지국이 동시에 서비스 가능한

상황에서더나은통신속도와안정성을보여주는것은당연하게보이지만,이를달

성하기위해서는잘짜여진트레픽관리가필요하다.

이 논문에서 우리는 LTE/mmWave 이중연결성 (dual connectivity) 환경을 바탕

으로, LTE와 mmWave라는 서로 다른 라디오 기술을 동시에 사용하기 위한 방법을

연구하였다.이두가지기술을동시에활용하기위해서,우리는트레픽라우팅기작

과 PDCP재배열기작을함께포함하고있는분할알고리즘을제안하고있다.또한

ns-3시뮬레이션을통해이알고리즘의성능을검증하였다.

주요어:이중연결성 (dual connectivity), LTE, mmWave,다경로동시전송 (multi-path

transmission)

학번: 2016-20936

17