LTE System Performance Optimization by Discard Timer Based PDCP Buffer Management

You might also like

You are on page 1of 7

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/254050689

LTE system performance optimization by discard timer based PDCP buffer


management

Article · December 2011


DOI: 10.1109/HONET.2011.6149800

CITATIONS READS

2 2,917

4 authors:

Umar Toseef Thushara Weerawardane


Nokia Bell Labs, Stuttgart General Sir John Kotelawala Defence University
44 PUBLICATIONS   250 CITATIONS    40 PUBLICATIONS   400 CITATIONS   

SEE PROFILE SEE PROFILE

Andreas Timm-Giel Carmelita Görg


Technische Universität Hamburg Universität Bremen
225 PUBLICATIONS   1,998 CITATIONS    299 PUBLICATIONS   2,439 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Modeling of Data Dissemination in OppNets View project

NOMAD (Integrated Networks for Seamless Service Discovery) View project

All content following this page was uploaded by Umar Toseef on 13 March 2016.

The user has requested enhancement of the downloaded file.


> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TOEDIT) < 1

LTE system performance optimization by RED


based PDCP buffer management
Umar Toseef1, Thushara Weerawardane1, Andreas Timm-Giel2, Carmelita Goerg1
1
TZI ComNets, University of Bremen, Bremen, Germany, {umr/tlw/cg}@comnets.uni-bremen.de
2
ComNets, TUHH, 21073 Hamburg Germany, timm-giel@tuhh.de

 Radio Link Control (RLC) layer, Medium Access Control


Abstract—Long Term Evolution (LTE) provides a new radio (MAC) layer and Physical (PHY) layer. All these four
technology that allows operators to achieve even higher peak protocol layers are part of the protocol stack in UE and
throughputs than HSPA+. One of the LTE system performance eNB. At the eNB side, each radio bearer has one PDCP
optimization point is the PDCP buffer management. PDCP entity which processes the IP packets in the user plane. The
buffers at eNB hold user data before it is scheduled by the eNB
PDCP layer mainly performs header
MAC scheduler for transmission over the radio interface. This
buffered data also contributes towards X2 traffic volume and
compression/decompression, security including data
required a certain inter-eNB handover completion time integrity/verification and ciphering/deciphering, and also
depending upon amount of data buffered in bearer buffers. A reordering during handovers. The main data buffers of Uu
suitable buffer management scheme is required to keep PDCP interface are located at the PDCP layer where data is held
buffer occupancy to a minimal level without a noticeable before its transmission to the destination UE. Each PDCP
degradation in user application performance. This work entity has its own PDCP buffers implemented on a per radio
suggests RED as buffer management scheme for PDCP bearer bearer basis. However, these buffers use common shared
buffers at eNB and provides a thorough study of its memory space in the eNB entity. In contrary to this, at the
performance for several real world scenarios. Through an UE side, the shared buffer capacity is allocated on a per-
intensive set of simulations effectiveness of proposed PDCP
user basis for the uplink transmission.
buffer management scheme in achieving an optimal PDCP
buffer occupancy is proved. Furthermore configuration As a high speed broadband network, LTE promises an
parameters are suggested to achieve optimum LTE system uninterrupted connectivity to packet data network even
performance using the proposed scheme. when UE is moving at high speed. However, in practice
when UE is moving from source network to target network,
Index Terms— LTE, PDCP buffer management, LTE there is definitely a connection break and make. During
handover performance optimization, RED buffer management inter-eNB handovers, the connection is broken from the
scheme, User QoE optimization source eNB and again made at target eNB, the data to UE is
buffered at the source eNB and forwarded to target eNB
INTRODUCTION over the X2 interface. There are two constituents of this
forwarded data: i) existing contents of all PDCP buffer data
T he roadmap of Next Generation Mobile Network
(NGMN) is to provide mobile broadband services to the
end-users. To make this happen, the 3GPP introduces Long
belonged to the UE and, ii) new incoming data packet from
aGW until (~80 ms delay) aGW switches the data delivery
Term Evolution (LTE) to ensure the ongoing path for this UE from source eNB to the target eNB. This
competitiveness of the 3GPP technology family. LTE traffic flow is termed as X2 forwarding data flow. Larger
introduces a new air interface and radio access network, the PDCP buffer contents (i.e. high buffer occupancy),
which provides much higher throughput and low latency, higher will be the X2 traffic volume. Large volume of X2
greatly improved system capacity and coverage than those traffic at a short period of time on one hand consumes
of the WCDMA systems. These improvements lead to expensive transport network bandwidth and on the other
increased expectations on the end-user quality of services hand it extends the handover delays leading poor end-to-end
over LTE as compared to existing 2G/3G systems. The application performance[6].
improvements impose much higher requirements on the Apart from its impact on handover performance an
transport network, as well: (i) as the LTE system is designed arbitrarily large memory space allocated for PDCP buffers
for Packet Switched (PS) traffic, PS based data services are could also impact end user application in another way and
expected to dominate in LTE; (ii) the significantly improved can be described as follows. Bigger the PDCP buffer space,
throughput and capacity of the LTE air interface will put longer will be the queuing delays for packets in air interface
more bursty load on the transport network; (iii) the higher congestion situations. This queuing delay could cause user
expectations on the responsiveness of interactive packet end-to-end delay grow so large that both real time
applications, and on voice quality increase the demand on and non-real time applications got adversely affected by the
lower packet delay in the transport network. higher layer protocols. Real time applications have strict
LTE radio interface, the interface between UE and eNB requirements on end-to-end packet delay which must be
consists of four main protocol layers to transfer the data followed to achieve acceptable user Quality of experience
between eNB and UE securely and with certain reliability. (QoE). For example conversation VoIP specifies mouth-to-
They are Packet Data Convergence Protocol (PDCP) layer, ear delay less than 150ms to achieve transparent
interactivity [1]. On the other hand non real time
applications do not impose strict packet end-to-end delay
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TOEDIT) < 2

but still user QoE for such applications is closely related to characteristics are modeled at the MAC layer in terms of the
packet end-to-end delay. For example, today’s most of non- data rates of individual user performance. For the UE
real time applications are TCP based and TCP throughput mobility, generic mobility model such as random directional
has direct relationship with TCP segment round trip time i.e. and random way points are used.

MSS
TCP throughpu t  (1)
RTT PLR
Where MSS is the maximum TCP segment size, RTT is
the TCP segment round trip time and PLR is the packet loss
rate[2].
In the light of above listed disadvantages, the goal of this
work is to devise a PDCP buffer management scheme which
keeps buffer fillings up to minimum required level in order
to achieve target radio throughputs while providing
optimum end-to-end application performance. Fig. 2. LTE Protocol Structure (user-plane)

LTE SIMULATION MODEL PDCP BUFFER MANAGEMENT SCHEME


The LTE simulator is implemented using the OPNET In order to achieve an optimum PDCP buffer occupancy a
simulation tool environment (version 15.0) [3] And shown buffer management scheme based on RED (Random Early
in Figure 1. It includes all basic E-UTRAN and Evolved Detection) is used. The reader is encouraged to refer [7] for
Packet Core (EPC) [4] network entities and protocols. an understanding of RED active queue management
Figure 1 depicts an example scenario with two eNBs, and a scheme. Fig. 3 shows an overview of packet discard
number of IP routers connecting the eNBs and the aGW in mechanism of proposed scheme which is based on buffer
the E-UTRAN network. The EPC network entities are occupancy level. As long as buffer occupancy is below the
represented by the aGW network entity. The remote node minimum threshold (minth) value, no packet is discarded.
represents an Internet server or any server/node that For buffer occupancy between minimum and maximum
provides the data or voice services for LTE users. threshold packets are discarded with a certain probability.
The LTE transport network is based on IP technology[5]. The probabilities of packet discard increase linearly from Pa
The user-plane transport protocols are implemented for both to Pb along with buffer occupancy increase from minimum
the S1 interface (i.e. the interface between the eNB and the threshold value to maximum threshold (maxth) value as
aGW) and the X2 interface (i.e. the interface between the shown in the figure 3. After maximum threshold value of
eNBs) according to 3GPP specification. The S1/X2 buffer occupancy the packet drop probability becomes 1.0.
interface mainly includes the user plane protocols such as At 100% buffer occupancy packets are discarded by drop
GTP, UDP, IP and layer 2 (L2) protocols. Ethernet is used tail mechanism.
as the layer 2 protocol in this simulation model.

Fig. 3. RED based PDCP buffer management scheme

The performance of RED based buffer management


Fig. 1. LTE Simulation Model in OPNET scheme is compared with the simple tail drop scheme. In
Fig. 2 shows the LTE user-plane protocol structure which is this scheme a maximum limit of per bearer PDCP buffer
developed within this LTE simulator. The protocols are capacity is imposed. Once buffer filling reach that
categorized into three groups: radio (Uu), transport, and maximum limit, all newly incoming packets belonged to this
end-user protocols. The radio (Uu) protocols include the bearer are discarded. Discarding action continues until some
peer to peer protocols such as PDCP, RLC, MAC and PHY of the data are accepted by MAC scheduler for transmission
between UE entity and eNB entity. The PDCP, RLC and which makes room for newly arriving data.
MAC (including air interface scheduler) layers are modeled
in detail according to the 3GPP specifications in this SIMULATION CONFIGURATION & SCENARIOS
simulator. But the PHY (physical) layer is not detailed TABLE I lists the LTE simulator configuration
modeled since our focus lies on the LTE access network. parameters used to get results for this work. This section has
However the effect of the radio channels and PHY been divided into four subsections. Each subsection covers a
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TOEDIT) < 3

discussion on a certain simulation scenario set. Each is increased to 60. Therefore, the amount of offered traffic
simulation scenario set provides a certain offered traffic load in scenario set#4 will be identical to scenario set#2
mixtures and traffic load levels in order to study the which is two times higher than that of scenario set#1 and
performance of proposed RED based buffer management scenario set#3.
TABLE II
scheme. The very first scenario of each simulation scenario
USER CONSTELLATION FOR SIMULATION SCENARIO SET 1 TO 4
set presents system performance when no buffer Number of UE/cell –
Number of UE/cell –
management scheme is used and an unlimited memory Simulation downlink uplink
space is assigned to PDCP buffers. This provides a baseline scenario set
QCI 8 QCI 9 QCI 8 QCI 9
for performance comparison with cases where tail drop or Set#1 20 HTTP 7 FTP None 3 FTP
RED scheme is deployed. Set#2 40 HTTP 14FTP None 6 FTP
TABLE I
15 HTTP + 5 HTTP +
LTE SYSTEM SIMULATOR, CONFIGURATION PARAMETERS Set#3 1 FTP 2 FTP
3FTP 4 FTP
User Profile Definition 30 HTTP + 10 HTTP +
Set#4 2 FTP 4 FTP
Number of active users per 30 6 FTP 8 FTP
cell
Number of cells per eNB 1 SIMULATION RESULTS
Number of users per eNB 30
FTP traffic data File size: constant 5M Byte This section discusses the simulation results for the
Web (HTTP) traffic data Number of pages per session: 5 pages aforementioned simulation scenarios. Simulation results of
Average page size: 100K Bytes the each of simulation scenario set are separately analyzed
Number of objects in a page: 1
by discussing the advantages and disadvantages.
TCP configuration at UE & New Reno, 64 Kbytes receive
FTP/HTTP server window size, MSS: 1460 bytes i. Simulation scenario set#1
User Mobility Model Random Direction (50km/h)
Fig. 4 shows buffer occupancy used by PDCP data at eNB.
Network Configuration It can be noticed that when shared PDCP buffer size is
Cell Bandwidth 10 MHz
Handover Disabled
virtually unlimited, the total buffer occupancy can grow as
MAC Scheduler Round Robin, QoS support enabled large as 350 Kbytes. However by limiting the buffer
S1 link capacity 100 Mbps (Ethernet 100BaseX) capacity available for each bearer to smaller values both the
Shared PDCP buffer capacity unlimited maximum and average buffer occupancy can be controlled.
Per bearer PDCP buffer Configurable It is clear that the maximum PDCP shared buffer capacity
capacity
RED parameters for PDCP minth: 33%, maxth: 100%
requirements for this particular scenario are 350 Kbytes/cell.
buffer management Pa: 0%, Pb: 5% 350 Kbytes is actually the sum of buffer occupancies by all
Simulation duration 2000 sec active bearers.
TABLE II shows the user constellation used in simulation
scenario set#1 to 4. These simulation scenarios can be
grouped according to their underlying user priority
assignment schemes.
a) Prioritizing HTTP users over FTP user
Simulation scenario set#1 make use of a typical priority
assignment scheme at Uu interface. In this scheme HTTP
users are prioritized over FTP users in order to provide
faster web access for the mobile users. The priority is
ensured by assigning QCI 8 class to the HTTP users and
QCI 9 class to the FTP users. In LTE simulator, the MAC
scheduler performs QoS aware scheduling where QCI 8
traffic is eligible to get five times higher resource share than
that of QCI 9 traffic. Simulation scenario set#1 creates only Fig. 4. Shared PDCP buffer occupancy for different per bearer buffer
a light congestion at Uu interface. However in order to capacity values. The figure shows CDF curves of total shared buffer
occupancy (bytes).
evaluate system performance under high congestion
situation, scenario set#2 is used. In this scenario set, the It is worth mentioning that maximum shared buffer
amount of offered load is doubled by increasing the number occupancy of 350 Kbytes is only valid for this particular
of users/cell. simulation scenario setup i.e. Uu interface congestion level,
relative priority of radio bearers, traffic model, traffic
b) Mixing HTTP and FTP users in same priority class
distribution of services etc. For example, TCP sends data in
In set#3 of simulation scenarios HTTP and FTP users are bursts which are more likely to be held in PDCP buffer
mixed together in priority classes QCI 8 & QCI 9. This is before transmission over Uu interface and hence requires
also one of the common priority assignment schemes for
large buffer capacity. Contrariwise real-time applications
user traffic. This represents real world scenario where, for
e.g. VoIP, Video etc. send data packets at regular intervals.
example, premium users are always assigned to QCI 8
These intervals are typically 5ms to 20ms and the
priority class and basic users fall in QCI 9 priority category.
In this scenario set there are 30 users per cell. In scenario probability is high that old packets are transmitted before
set#4, similar to scenario set#2 but the number of users/cell the arrival of new packets due to high priority they process
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TOEDIT) < 4

at the scheduler compared to non-real time traffic. however number of packet drops rise to a very high number.
Therefore, it is less likely that VoIP data will be held in Discarding one PDCP packet leads to a loss of one TCP
PDCP buffer in large amounts and leads to larger buffer segment and that has to be retransmitted at TCP layer to
utilization. Relative priority of bearers at MAC scheduler guarantee the reliable error free transmission. Higher the
also plays an important role in deciding the PDCP buffer packet loss rate, more the TCP retransmissions occur and in
occupancy. The higher priority traffic experiences a shorter turn longer will be the file download time. Furthermore, in
waiting time before transmission over Uu interface as severe situations where packet drop rate grow very high
compared to that of lower priority traffic flows. In this way some TCP connections may not be able to recover leading
high priority traffic flows are less likely to have high PDCP to connection abort. However TCP achieves shorter round
buffer occupancy compared to low priority traffic such as trip time when PDCP buffer limitation is imposed. This can
FTP. Based on these facts it can be inferred that requirement be seen in Fig. 5 as well as in TABLE III. From equation 1
of maximum PDCP buffer occupancy could change if any it is evident that reduction in TCP segment delay increases
alteration appears in number of users, traffic mixture, air TCP throughput. Therefore limiting of maximum shared
interface congestion level etc. This will also be observed in buffer capacity on one hand degrades TCP throughput due
simulation results of scenarios discussed later in this section. packet loss but on the other hand it enhances TCP
TABLE III throughput by reducing TCP segment delay. The final gain
SIMULATION SCENARIO SET #1, SIMULATION RESULT STATISTICS
or loss in TCP performance is decided by combing the
Per Mean Total Mean Mean FTP Mean
bearer PDCP PDCP TCP file HTTP
impact of the two factors. In current simulation setup the
PDCP buffer packet segment download page resulted impact can be judged by the HTTP and FTP file
buffer space drops delay – time (sec) download download time as given in TABLE III. The best HTTP and
capacity usage FTP DL time (sec)
(KB) (KB) (msec)
FTP application performance is achieved when no limitation
unlimited 61.7 0 115 9.41 0.54 is imposed on per bearer PDCP buffer size and the worst
Tail drop
application performance is observed when a limit of 30
100 61.7 0 115 9.41 0.54 Kbytes is enforced. The drawback of having unlimited
50 48.0 994 094 10.01 0.54 shared PDCP buffer capacity is very high PDCP buffer
30 29.5 5,078 051 11.21 0.53
occupancy which is undesirable. When PDCP buffer
RED capacity is limited to 50 Kbytes the buffer occupancy can be
100 43.7 653 082 10.10 0.53
50 30.9 2,033 058 11.43 0.56
reduced approximately by 22% at the cost of ~6%
30 24.1 5,080 044 13.28 0.70 degradation in FTP application performance. Further
reducing PDCP buffer capacity to 30 Kbytes can reduce
buffer occupancy by 52% at the cost of 19% in FTP
application performance degradation. Aforementioned
performance figures holds when tail drop scheme is used.
With RED scheme and 100 Kbytes per bearer buffer
capacity 30% reduction in total shared buffer occupancy is
achieved at the cost of 7% degradation in FTP application
performance. The performance figures imply the superiority
of per bearer based RED over tail drop scheme in this
particular case. Furthermore with RED scheme and 50
Kbytes per bearer buffer capacity 50% reduction in total
shared buffer occupancy is achieved at the cost of 21%
increase in FTP file download time.
Fig. 5. TCP segment delay in downlink direction for different per bearer ii. Simulation scenario set#2
buffer capacity values. The figure shows the CDF curves of TCP segment
delay for FTP downlink
In this Simulation scenario set, the traffic mixture is
TABLE III shows several simulation result statistics identical to that of simulation scenario set#1. However the
including total PDCP packet drops due to buffer overflow. amount of offered traffic load has just been doubled by
This number represents the accumulated count of PDCP twofold increase in number of users per application. Due to
packet drops during the whole simulation time of 2000 sec. severe congestion created by larger offered traffic volume
When per bearer PDCP buffer capacity is not limited, no PDCP buffer occupancy is expected to increase accordingly.
packet drop can be observed. Further zero packet drops is In the following performance of tail drop and RED buffer
also observed when per bearer PDCP buffer capacity is management schemes are analysed under this user
limited to 100 Kbytes along with the tail drop scheme. This constellation.
implies that per bearer PDCP buffer capacity requirements Fig. 6 show total shared PDCP buffer occupancy in the
are not larger than 100 Kbytes for this particular scenario. form of CDF curves. In contrast of Fig. 4, the CDF curves
On further reducing per bearer buffer capacity to 50 Kbytes in Fig. 6 have steep slope which indicates high utilization of
provides 22% reduction in total shared buffer occupancy buffer capacity. It can be noticed that when no limitation on
along with several hundred of packet drops. When per PDCP buffer capacity is imposed the buffer occupancy goes
bearer buffer capacity is limited to 30 Kbytes it promise as high as 2.1 Mbytes. Comparing no buffer limitation cases
52% reduction in total shared PDCP buffer occupancy in Fig. 4 & Fig. 6 reveals that two times increase in offered
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TOEDIT) < 5

traffic volume has produced six fold increase in maximum reduction in PDCP buffer occupancy as compared to the
buffer occupancy. In previous simulation scenario set 100 unlimitted per bearer buffer capacity case. Further reducing
Kbyte per bearer PDCP buffer capacity did not produce any per bearer buffer capacity to 50 Kbytes the shared PDCP
packet drops. However in current simulation scenario set buffer occupancy can be reduced by 70% at the expense of
limitation of 100 Kbytes per bearer buffer capacity yields only 2% increase in FTP file download time as compared to
much less buffer occupancy as compared to unlimited buffer unlimited per bearer buffer capcity case. This analysis
capacity case. This indicates a high possibility of packet indicates that buffer management is of much more
drops due to buffer limitation of 100 Kbytes per bearer. This importance when air interface is congested as compered to
intutive inference is confirmed by the simulation results non congested case.
presented in TABLE IV which shows several thausands of
packet drops for 100 Kbytes per bearer buffer limitation
case. TABLE IV indicates that reducing per bearer buffer
capacity from 100 Kbytes to 50 Kbytes reduces buffer
occupancy by ~45%. Likewise reducing per bearer buffer
capacity by 70% from 100Kbytes to 30 Kbytes also reduces
the buffer occupancy by 68%.

Fig. 7. TCP segment delay in downlink direction different per bearer buffer
limitation values. The figure shows the CDF curves of TCP segment delay
for FTP downlink
RED case with 30 Kbytes per bearer buffer capacity
outperforms tail drop scheme by providing ~84% reduction
in PDCP buffer occupancy at the cost of only 5.6% increase
in FTP download time when comapred to unlimited per
Fig. 6. Shared PDCP buffer occupancy for different maximum buffer size bearer buffer capcity case. Tail drop with 30 Kbytes per
values. The figure shows CDF curves of shared buffer occupancy (bytes). bearer buffer capacity could only achieve 79.5% reduction
TABLE IV in shared PDCP buffer occupancy at the cost of 7.2%
SIMULATION SCENARIO SET #2, SIMULATION RESULT STATISTICS degradation in FTP performance. This analysis shows that
Per Mean Total Mean Mean Mean RED buffer management scheme is superior choice when
bearer PDCP PDCP TCP FTP file HTTP
PDCP buffer packet segment download page
very low PDCP buffer occupancy is desirable or in
buffer space drops delay – time (sec) download otherword at severe congestion phases of the Uu interface.
capacity usage FTP DL time (sec) A good balance between PDCP buffer occupancy
(KB) (KB) (msec)
reduction and application performance degradtion can be
unlimited 1450 0 2,189 113.17 0.68
achieved either by RED with 100 Kbytes or tail drop with
Tail drop
100 932 4,442 1,342 112.12 0.68
50 Kbytes per bearer limitation choice.
50 508 11,941 763 115.36 0.68
30 296 20,170 475 121.32 0.68
iii. Simulation scenario set#3
RED
In this simulation scenario set FTP and HTTP users are
100 528 6,754 746 114.21 0.68 mixed together in priority classes QCI 8 and QCI 9. The
50 323 14,810 520 117.28 0.78 presence of FTP user in higher MAC priority class of QCI8
30 227 24,673 361 119.56 0.94 would increase offered traffic load belonged to QCI8 class
On the other hand reduction buffer occupancy brings due to high priority at MAC scheduling at the Uu interface
great improvement in TCP segment delay for downlink. Fig. and hence QCI9 traffic will be left with less resources
7 shows the CDF curves of TCP segment delay for all available at air interface as compared to that of scenario
scenarios in this simulation set. Compared to Fig. 5 a huge set#1. FTP QCI9 users with reduced air interface throughput
increase in maximum TCP segment delay can be observed. would in turn cause high PDCP buffer occupancy.
Though limitation of buffer capacity improves TCP segment TABLE V lists the important result statistics obtained for
delay, the magnitude of values is still very high. TABLE IV simulation of scenario set#3. A slight difference in FTP file
shows that 50% decrease in per bearer PDCP buffer download time for QCI8 and QCI9 traffic points to a mild
capacity (from 100 Kbytes to 50 Kbytes) also brings ~50% congestion at air interface. Tail drop case with 100 Kbytes
improvement in TCP segment delay. Similarly 70% cut-off still seems enough to meet all requirements of PDCP data as
per bearer PDCP buffer capacity brings ~68% reduction in indicated by zero packet drops. It can be observed that
TCP segment delay. performance degradation of QCI8 applications is relatively
TABLE IV shows that 100 Kbytes per bearer buffer less severe as compared to that of QCI9 applications when
limitation case exhibits a win-win situation i.e. slight lower buffer limitation is enforced. The rationale behind this
improvement in FTP performance along with ~36% behavior is the higher buffer occupancy and packet drops of
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TOEDIT) < 6

QCI9 traffic than that of QCI8 traffic. The PDCP buffer RED are listed in TABLE I.
occupancy of a radio bearer is determined by the difference TABLE VI
SIMULATION SCENARIO SET #4, RESULT STATISTICS
of incoming data rate from transport network and outgoing
data rate to Uu interface. Therefore if radio bearers with
different priority exist in a cell then radio bearers with lower
transmit priority are naturally expected to have larger PDCP
buffer occupancy as compared to that of bearers with higher
transmit priority. The large PDCP buffer occupancy hints to
high TCP segment delays and probability of packet loss
leading to performance degradtion.
RED case with 100 Kbytes per bearer buffer limitation
can be considered as the optimum case which reduces PDCP
buffer occupancy by 37% as well as improves HTTP
QCI8/QCI9 and FTP QCI8 performance at the expense of CONCLUSION
1.8% increase in FTP QCI9 file download time.
TABLE V An optimal PDCP buffer occupancy at eNB can reduce
SIMULATION SCENARIO SET #3, SIMULATION RESULT STATISTICS the X2 traffic volume and the required handover procedure
time, thereby achieving gains in user application
performances. This work investigates LTE system
performance optimizations using a PDCP buffer
management scheme at eNB. For this purpose, a buffer
management scheme based on Random Early Detection
(RED) is introduced and implemented in OPNET based
LTE system simulator. With the help of several different
simulation scenarios it is shown that limiting PDCP buffer
capacity without buffer management leads to packet drops
in burst i.e. the tail drop phenomenon. Though lower PDCP
iv. Simulation scenario set#4
buffer occupancy is achieved by the tail drop scheme, it is
The simulation scenario set#4 has traffic mixture identical
not always the optimal choice. On the other hand, when the
to scenario set#3 but the number of users for each
proposed RED based buffer management which is TCP
application are increased two times in order to bring twofold
friendly random packets discard algorithm is used, a good
increase in offered traffic load. TABLE VI can be referred
balance between PDCP buffer occupancy and application
for the simualtion result s of this scenario set.
performance can be attained. The study of simulation results
The result statistcs show that the increase of offered traffic
with different offered traffic loads and traffic mixtures
has resulted a drastic increase in both mean PDCP shared
prove the applicability and usefulness of the RED based
buffer occupancy and toal number of PDCP packet discards.
buffer management scheme in achieving LTE system
Compared to previous simulation scenario set#3 following
performance optimization goals. A practical range of tuning
observations can be made:
parameter values is also proposed to ease the system
- Mean PDCP buffer occupancy is almost doubled for
performance optimization task when using RED based
unlimited per bearer buffer case
PDCP buffer management scheme.
- Per bearer buffer limitation of 100 Kbytes with tail
drop is not sufficient to hold PDCP data without
REFERENCES
overflow
[1] Recommendation ITU-T G.714, International telephone connections
- PDCP buffer occupancy is almost unchanged for all and circuits – General Recommendations on the transmission quality
cases except tail drop with 100 Kbytes per bearer for an entire international telephone connection, Approved in May
buffer limitation 2003
[2] Matthew Mathis and Jeffrey Semke and Jamshid Mahdavi and Teunis
- The number of packet drops has increased to very high Ott, The Macroscopic Behavior of the TCP Congestion Avoidance
values compared to scenario set#3. Algorithm, Computer Communications Review, volume 27, number 3,
- There is a huge difference of user application July 1997
performance (i.e. HTTP/FTP download time) between [3] OPNET official website, http://www.opnet.com, as accessed in June
2011
QCI8 and QCI9 traffic which hints severe congestion [4] 3GPP technical Report TR 25.913. Requirements for Evolved UTRA
at air interface. and UTRAN. V 2.1.0, June 2005.
In this congested air interface situation RED with 100K [5] Stefania Sesia, Issam Toufik, and Matthew Baker. LTE - The UMTS
Long Term Evolution, From Theory to Practice. John Wiley and Sons
bytes per bearer buffer limitation provides an optimum Ltd, 1st edition, 2009
performance by reducing the PDCP buffer occupancy to [6] 3GPP technical specification TS 23.40, General Packet Radio Service
47% and reducing FTP QCI9 file download time by 8.3%. (GPRS) enhancements for Evolved Universal Terrestrial Radio
From the analysis of simulation results presented in all Access Network (E-UTRAN) access, V8.11.0 September 2010.
four subsections it can be recommended that RED based [7] Floyd, S., and Jacobson, V., Random Early Detection gateways for
Congestion Avoidance, IEEE/ACM Transactions on Networking, V.1
buffer management scheme with 100 Kbytes per bearer N.4, August 1993, p. 397-413
buffer limitation can be used to achieve optimum LTE
system performance. Other configuration parameters of

View publication stats

You might also like