04_RA47054EN16AGLA0_Throughput_Optimisation.pdf

Throughput Optimization Slide 1

NokiaEDU Throughput Optimization LTE Optimization Principles [FL16A] Module 04

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

1

Throughput Optimization Slide 2

Copyright and confidentiality

The contents of this document are proprietary and confidential property of Nokia. This document is provided subject to confidentiality obligations of the applicable agreement(s). This document is intended for use of Nokia’s customers and collaborators only for the purpose for which this document is submitted by Nokia. No part of this document may be reproduced or made available to the public or to any third party in any form or means without the prior written permission of Nokia. This document is to be used by properly trained professional personnel. Any use of the contents in this document is limited strictly to the use(s) specifically created in the applicable agreement(s) under which the document is submitted. The user of this document may voluntarily provide suggestions, comments or other feedback to Nokia in respect of the contents of this document ("Feedback").

2

Such Feedback may be used in Nokia products and related specifications or other documentation. Accordingly, if the user of this document gives Nokia Feedback on the contents of this document, Nokia may freely use, disclose, reproduce, license, distribute and otherwise commercialize the feedback in any Nokia product, technology, service, specification or other documentation. Nokia operates a policy of ongoing development. Nokia reserves the right to make changes and improvements to any of the products and/or services described in this document or withdraw this document at any time without prior notice. The contents of this document are provided "as is". Except as required by applicable law, no warranties of any kind, either express or implied, including, but not limited to, the implied warranties of merchantability and fitness for a particular purpose,

RA47054EN16AGLA0

are made in relation to the accuracy, reliability or contents of this document. NOKIA SHALL NOT BE RESPONSIBLE IN ANY EVENT FOR ERRORS IN THIS DOCUMENT or for any loss of data or income or any special, incidental, consequential, indirect or direct damages howsoever caused, that might arise from the use of this document or any contents of this document. This document and the product(s) it describes are protected by copyright according to the applicable laws. Nokia is a registered trademark of Nokia Corporation. Other product and company names mentioned herein may be trademarks or trade names of their respective owners.

© Nokia 2016

RA47054EN16AGLA0

2

Throughput Optimization Slide 4

Module Objectives • After completing this module, you will be able to:

• Describe features related DL and UL throughput and their impact. • Explain the impact of inter-cell interference • Discuss the relevance of UDP measurements

• Describe TCP Optimization principles • Give an overview about FTP troubleshooting

4

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

3

Throughput Optimization Slide 5

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

5

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

4

Throughput Optimization Slide 6

Traffic throughput PDCP layer throughput matches closest to application layer throughput. - Calculated with 1 second averaging => typically KPIs are showing lower throughput than experienced by users. - Averaging results over several cells further impacts on the values, as cells with no traffic are counted in with zero throughput. • PDCP mean DL - LTE_5292d Average PDCP Layer Active Cell Throughput DL, kbps • PDCP mean UL - LTE_5289d Average PDCP Layer Active Cell Throughput UL, kbps • PDCP peak DL - LTE_291a Maximum PDCP Throughput DL, Mbps • PDCP peak UL - LTE_288a Maximum PDCP Throughput UL, Mbps

Example values only!! Network average from different networks (A, B, C,…)

KPI AV PDCP active cell DL/Cell (LTE_5292)

AV PDCP active cell UL/Cell (LTE_5289)

6

A

B

C

….

X

Y

Z

21206 kbits

20452 kbits

20067 kbits

….

6994 kbits

5190 kbits

5055 kbits

3754 kbits

3054 kbits

2436 kbits

….

511 kbits

502 kbits

495 kbits

RA47054EN16AGLA0

© Nokia 2016

Improvements in RL50: New counters for Active cell time ACTIVE_TTI_UL (M8012C89) ACTIVE_TTI_DL (M8012C90) Communication with R&D Jan 31, 2012: What is currently implemented is a counter that gives data throughput directly PDCP_DATA_RATE_MAX_DL (M8012C25) PDCP_DATA_RATE_MAX_UL (M8012C22) PDCP_DATA_RATE_MEAN_UL (M8012C23) PDCP_DATA_RATE_MEAN_DL (M8012C23) Max counter can be used to calculated the max throughput somewhere in the network. This is fine. But mean rate has some limitation that are very difficult to handle. Throughput can be calculated only as cell throughput, but not as user throughput. Throughput is calculated per second only. Download of 1 Mbit (e.g surfing in internet) is counted as 1Mbit/sec even if the real throughput took only 100 ms. -> Value of throughput is low. Throughput counters are ok on cell and granularity base but difficult to handle on BTS or network level. NetAct is using all granularities for averaging: One cell without traffic give 0, one eNB with 1 cell with 10 Mbit/s and 2 cell without traffic give 3,33 MBit/s. But we want to have user throughput. With reporting Suite we can handle this, but with problems in performance. We have KPI for Average cell throughput (also on BTS or network level) but they need to be calculated with raw counters (as reporting Suite is doing) LTE_5289b Average PDCP Layer Active Cell Throughput UL LTE_5292b Average PDCP Layer Active Cell Throughput DL

RA47054EN16AGLA0

5

Throughput Optimization Slide 7

Active cell time Counters for improving U-plane throughput measurements in UL and DL Active cell time ACTIVE_TTI_UL (M8012C89) ACTIVE_TTI_DL (M8012C90)

Number of TTIs in UL with at least one UE scheduled to transmit user plane data (no Cplane/signaling/control information considered) Trigger event: every TTI, in which at least one UE is scheduled to transmit/receive user plane data. Use case: This KPI enables new (more accurate) way of calculating the active cell throughput:

TTI w ith users scheduled to transmit User Plane data TTI w ithout scheduled users No scheduled users in the cell

Data is transmitted

ACTIVE_TTI_* = Count( PDCP_SDU_VOL_ * =

7

Σ

8* (PDCP_SDU_VOL_DL)

PDCP Layer Active Cell Time (ms) Throughput DL

(LTE_5292d) =

PDCP Layer Active Cell Throughput UL

(LTE_5289d) =

ACTIVE_TTI_DL

) [#] [bytes]

RA47054EN16AGLA0

8* (PDCP_SDU_VOL_UL) ACTIVE_TTI_UL

© Nokia 2016

RA47054EN16AGLA0

6

Throughput Optimization Slide 8

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

8

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

7

Throughput Optimization Slide 9

RF Peak Tput Under Neighbor Cell Interference • Measuring peak MIMO dual-stream throughput in the field can be tricky because of interference • An idle cell produces common channel + RS interference to impact peak throughput  need to find good interference-free measurement spot.

Inter-site cell border, non-frame synchronized cells

Intra-site cell border, framesynchronized cells

9

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

8

Throughput Optimization Slide 10

Impact on Peak Tput from Idle Neighbor Cell Interference • Measurement example #1, Samsung terminal, 20MHz • inter-site and intra-site neighbor are unloaded (no PDSCH traffic) All neighbor cells attenuated 50dB

PHY tput, CINR, RSRP 120

Inter-site interference, adjacent site cell about 5 dB weaker RSRP than serving cell

Intra-site interference, adjacent cell about 5 dB weaker RSRP than serving cell

-60

-70

80 -80

60

-90

RSRP dBm

PHY tput Megabits/sec, CINR dB

100

-50

Data Average of Phy DL TP(Mbps) Average of SCell-CINR Average of SCell-RSRP

-100 40 -110 20 -120

06/11/2010 09:53:02.801

0 06/11/2010 09:54:45.317

-130 06/11/2010 09:56:29.840 Time

10

Intra-site neighbor frame-synced, no RS interference

RA47054EN16AGLA0

06/11/2010 09:58:14.347

All neighbor cells attenuated 50dB © Nokia 2016

RA47054EN16AGLA0

9

Throughput Optimization Slide 11

TD-LTE, impact of idle mode interference on tput • UE FTP downloading in the middle of two sectors of the same site, RSRP from both cells ~ -70dBm

• First the second cell is off (rebooting),

34Mbps vs 15Mbps

Neighbor cell switched on

SINR

tput

Serv RSRP

then comes on-air idle (only common channels transmitted)

11

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

10

Throughput Optimization Slide 12

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL proactive scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

12

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

11

Throughput Optimization Slide 13

Optimizing 2x2 Dynamic Adaptive MIMO Switching SM

CQI

Upgrade Switch : If mimoCQI > mimoSmCqiThUpOL and mimoRANK > mimoSmRiThUpOL

SM

mimoOlCqiThU mimoOlCqiThD

RI

Downgrade Switch:: If mimoCQI 1000.0

Shaping on*

Shaping off

*throughput fluctuations seem to be driven mainly by RF conditions

20

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

19

Throughput Optimization Slide 21

Traffic Shaping not Activated in Transport Network - Whenever feeding a low bandwidth box with high bandwidth input, check shaping in the transport chain • Especially if the tx buffer in the low bw box is small packet loss can occur  reduces TCP throughput! Example: A-2200 feeding FlexiPacket A2200

315Mbps 1Gbps

A2200/A1200 FPR ODU

Activate shaping here to reduce burstiness

Box X

FPR ODU

Packet loss here due to too small tx buffer relative to data burstiness

Data transfer direction Generalization: thick pipe

Box Y

thin pipe

Box Z

Be alert for packet loss here !

21

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

20

Throughput Optimization Slide 22

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

22

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

21

Throughput Optimization Slide 23

DL performance example features

Other features affecting DL performance RL60 • DL CA 40 MHz • Inter frequency load balancing

FL16 • DL SU 4x4 MIMO • CA enhancements • Liquid cell

RL70 • eICIC • DL interference shaping • DL CA 3 CC – 40 MHz • Inter frequency Load equalization

FL15A • DL CA 3 CC – 60 MHz • Flexible SCell Selection • eICIC for HetNet eNodeB Configurations • eICIC Enhancements - micro • Additional CA Band Combinations – II • Inter-eNodeB CA for 2 Macro eNodeBs

23

RA47054EN16AGLA0

FL16A • 256 QAM in DL • CA-aware Idle Mode Load Balancing • RSRQ-based A5 • FDD DL CA 4CC

© Nokia 2016

RA47054EN16AGLA0

22

Throughput Optimization Slide 24

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

24

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

23

Throughput Optimization Slide 25

Uplink performance – balance between user vs. cell performance Optimization goals: • Maximize peak throughput • Maximize average cell throughput • Minimize inter cell interference while maintaining cell edge throughput • Minimize scheduling delay Related features: • UL power control setting - Closed loop: SINR and RSSI target window. IAWPC - Open loop: P0 and alpha factor. • UL link adaptation - LA algorithm: MCS vs. allocated PRBs. • UL scheduler - Proactive UL scheduling - UL scheduler FD type: RoundRobing/Exhaustive/Mixed - Scheduling method of the UL scheduler: CUS/IAS/CAS

25

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

24

Throughput Optimization Slide 26

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput • How to troubleshoot packet loss related TCP tput problems?

26

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

25

Throughput Optimization Slide 27

Simulation Setup UL Name

Configuration parameters

Cluster

3GPP ISD500m 7 sites / 21 cells – wrap-around

Operation band

FDD 2000 MHz

Frequency bandwidth

10MHz (50PRB)

Antenna

3GPP 14 dBi

Tx Power

max 22.5 dBm

Penetration loss

20dB (outdoor to indoor)

No. of UEs

50 UEs per cell, randomly distributed

Mobility

3km/h speed

Traffic model

FTP traffic: 10 MB file

Simulation time

200 s

Slow Fading Fast Fading

On, 8 dB sigma On (Typical urban)

UL Power control

See slide no 2.

Scheduler

TDS: PF (Proportional Fair) FDS: Channel aware + Round Robin

MCS

QPSK/16QAM (MCS0 to MCS24)

System load

10% / 25% / 50% / 100%

• load realization • limitation of max # users scheduled per TTI • limitation of max # PRBs per user • total # PRBs per cell • max # avail PRBs: 50

27

5 5

RA47054EN16AGLA0

10% 1 6 12

25% 2 5 25

50% 5 50

100% -

© Nokia 2016

RA47054EN16AGLA0

26

Throughput Optimization Slide 28

Very similar results for all CL settings. 9.6 – 9.9 Mb/s

Very clearly the CLPC outperforms the OLPC in terms of cell throughput

Both OL settings achieve ~20% worse mean cell TP than CL settings. (7.3 – 7.9 Mb/s)

28

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

27

Throughput Optimization Slide 29

Only OL settings allows transmit for cell edge UEs. -80/0.8 better for IS500m than -100/1.0

With large CLPC window cell edge tput much better i.e. less inter-cell interference No transmission on the cell edge at all when CL PC is applied.

Trade off between mean cell TP and cell edge UE TP visible.

29

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

28

Throughput Optimization Slide 30

Power control and Link adaptation - Selected UL PC algorithm has clear impact on the UL throughput as well as the link adaptation mode and LA parameters

- The UL CLPC with large SINR window provides the best throughput per RB regardless of the selected LA method • Large SINR window provides good throughput while reducing the inter cell interference significantly • eUlLa protects the PUSCH BLER maintaining it at 10% level while SlowAmcOllaATB has slightly higher BLER

30

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

29

Throughput Optimization Slide 31

LTE1336: Interference aware UL power control Interference in LTE UL plays critical role determining the throughput UL Power control illustration

UL interference comes from neighboring cell UEs -

eNodeB1

It cannot be predicted since we do not know how the UEs will be distributed

eNodeB2

UL interference

Contributions to interference are most prominent from cell-border UEs

As a result the role of PC is to provide the required SINR while controlling at the same time the interference caused to neighboring cells

Looking at the UL SINR formula… SINR =

S I+N

=

Affected UE

Cell- center UE

Useful signal is already close to maximum

S Σ(I1,I2, … In)

Noise N is negligible in interference limited scenarios: I >> N

… it is clearly visible that it’s better to minimize the interference than improve the useful signal if we want to get better SINR

31

Cell-edge UE

RA47054EN16AGLA0

Cell edge users inject as much interference to neighboring eNB as to their serving eNB. Therefore contributions to interference in adjacent cells are most prominent from cell-border UEs

Minor UL interference to eNodeB1

© Nokia 2016

So if the interference is in the name of the feature let’s recap some information on it…

RA47054EN16AGLA0

30

Throughput Optimization Slide 32

LTE1336: Interference aware UL power control Interference-aware UL-PC deals with various SINR values which might easily be confused

UL SINR • UE-specific SINR of the UL transmission, received at the eNB • UL SINR is one of the inputs into the control loop, and is compared to a variable threshold, SINR_Target

Own cell

Own UE

Other cells

Pown ,PL

DL SINR

Pother ,PL’

SINR Target

DL CQI

SINR Target • UE-specific target value for the UL SINR calculated by LTE1336 • IAwPC calculates the transmit power-up/down commands to its UEs such that UL SINR is kept as close as possible to SINR_Target • Depends (among other data) on a ratio PL’/PL of pathlosses at the UE’s location, which the eNB derives from DL SINR (CQI)

+

UL Signal

DL SINR

RSSI

• DL SINR received at the UE’s location • It is a measure of the ratio of the pathloss from the UE towards the eNB representing the most prominent DL interference source, and the pathloss towards the own eNB • Reflected in the UE’s CQI reports which are used in SINR_Target calculation

32

Power command

-

UL Interfer ence

RA47054EN16AGLA0

UL SINR

UL Interference

© Nokia 2016

RA47054EN16AGLA0

31

Throughput Optimization Slide 33

LTE1336: Interference aware UL power control

kbps

Mean PDCP Tput (CLPC vs. IA ULPC) 7000 6000 5000 4000 3000 2000 1000 0

PDCP Cell Tput Level: NE 14.04.2014 10:00 - 14.04.2014 10:45 14.04.2014 10:45 - 14.04.2014 11:00 14.04.2014 11:15 - 14.04.2014 11:45

CL PC IAW PC disabled

14.04.2014 10:00

14.04.2014 10:15

14.04.2014 10:30

LTE_5289c/Average PDCP Layer Active Cell Throughput UL

IAW PC enabled

14.04.2014 10:45

14.04.2014 11:15

14.04.2014 11:30

LTE_5292c/Average PDCP Layer Active Cell Throughput DL

FL16A: LTE_5289d, LTE_5292d

Average UL SINR (OLPC vs. IA ULPC) Average UL SINR OLPC: 6.42dB IA ULPC: 8.53dB

14.000 %

12.000 % 10.000 % 8.000 % 6.000 % 4.000 %

OLPC

2.000 %

IA ULPC

0.000 %

33

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

32

Throughput Optimization Slide 34

LTE1336: Interference aware UL power control Average UL SINR (CLPC vs. IA ULPC) 14.000 %

Average UL SINR CLPC: 7.95dB IA ULPC: 10.95dB

12.000 % 10.000 % 8.000 % 6.000 %

4.000 %

CLPC

2.000 %

IA ULPC

0.000 %

Average UL MCS (OLPC vs. IA ULPC) 12.000 %

Average UL MCS OLPC: 7.28 IA ULPC: 10.16

10.000 % 8.000 % 6.000 %

OLPC

4.000 %

IA ULPC

2.000 % 0.000 %

34

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

33

Throughput Optimization Slide 35

Conclusion of the feature LTE1336 performance (CLPC vs. IA ULPC) Capacity gain 

ULTPIAULPC / ULPRBIAULPC  ULTPOLPC / ULPRBOLPC ULTPOLPC / ULPRBOLPC

31 % gain

220.00

This gain is only indicative of these tests, results from other networks will vary!

200.00 180.00 160.00

CLPC IA ULPC M8012C23/PDCP Throughput UL Mean (kbps) 1728.81 2162.98 M8011C24/UL PRB utilization per TTI Mean (%) 26.53 25.28 M8001C217/Average number of available PRBs 41.12 41.14 per TTI on PUSCH UL tput per PRB (kbps) 158.45 207.96 capacity gain (%) 31 %

140.00 120.00 100.00

CLPC IA ULPC

80.00

60.00 40.00 20.00 0.00

PM counters (raw data) and formulas:

35

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

34

Throughput Optimization Slide 36

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

36

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

35

Throughput Optimization Slide 37

UL Proactive scheduling Benefits: • Improves RTT by decreasing UL scheduling delay

Drawbacks: • Increases UL interference • Increases PDCCH usage

The feature is improves end-user performance in low loaded networks. When UL load increases, feature settings should be optimized.

37

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

36

Throughput Optimization Slide 38

Proactive scheduling – case example (1/3) - The UL throughput suffers from inter cell interference as can be seen from the interference analysis on TTI level: • Several consecutive TTIs are having increased interference levels due to the scheduling done under the interfering cell

• It should be noted that the proactive scheduling set to 50ms increases each of the allocations for each UE for 50 consecutive TTIs causing increased interference levels

Proactive scheduling 50ms

Base level is higher since somebody is transmitting in UL (e.g. for 50ms) in adjacent cells in a cluster

38

RA47054EN16AGLA0

Much longer duration of interference simply because of proactive scheduling timer of 50ms

© Nokia 2016

RA47054EN16AGLA0

37

Throughput Optimization Slide 39

Proactive scheduling – case example (2/3) - When proactive UL scheduling is turned off the impact of inter cell interference is greatly reduced as can be seen from the graph below • Indicating that most the allocations per UEs are just 1ms in length most applications people are using is very short burst which requires only small number of TTIs (e.g. 1 to a few)

Proactive scheduling off

Base level is lower since interference is really bursty in time domain

Almost continuous allocation of PRBs e.g. video call or streaming call

39

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

38

Throughput Optimization Slide 40

Proactive scheduling – case example (3/3) - When UL Proactive Scheduling is turned off UL App. Throughput per RB is increased 67% (drive testing at around -95dBm) • UL Interference level is reduced 5.8dB(5.3%) and PUSCH SINR is increased 5.0dB compared to Proactive scheduling with 50ms data.

- PUSCH RSSI is sustained but SINR is increased due to the interference reduction : 16dB 19.9dB, 24.4% - UL PRB utilization : 1310.7, 17.7 %

- PUSCH BLER is decreased : 8.04%3.18%, 63%

40

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

39

Throughput Optimization Slide 41

Uplink scheduling algorithms •

TTI PRB Map shows – CUS does not care about “Interference” – IAS cares about “Interference” only – CAS tries to trace channel condition actively

[PRB allocation – CUS]

With this behavior, CAS is expected to have better performance than IAS/CUS But under certain conditions only i.e. low speed UE channel conditions

[PRB allocation – CAS]

[PRB allocation – IAS]

Actively move PRB area (The first PRB is located on the most preferred segment)

Rarely move PRB area in the most preferred segment’s edge.

: Interfered Area

41

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

40

Throughput Optimization Slide 42

Uplink scheduling algorithms Live Network Testing Experiences - CAS shows better performance than IAS/CUS when lower AWGN SINR and moderate fading is inserted • In case of excellent RF condition, i.e., where MCS > 21 with CUS, CAS hardly has gain versus IAS/CUS as there is no room for improvement by channel aware scheduling - CAS/IAS tends to have higher gain when narrower PUSCH band is assigned to a single UE (gain for CAS has stronger impact by the #allocated PRBs) • When UL traffic is high enough with many UEs using VoLTE and PSVT, CAS/IAS shows higher gain than CUS - CAS/IAS Gain is degraded when fast moving model applied

• CAS shows better performance only with EPA5 channel model (Pedestrian, 3Km/h)

42

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

41

Throughput Optimization Slide 43

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

43

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

42

Throughput Optimization Slide 44

UL performance features Features to improve UL performance in RL50:

•

-

4RX UL Diversity with MRC.

-

4RX UL Diversity with IRC.

Features to improve UL performance in RL60: -

•

•

•

44

Fast ULA

Features to improve UL performance in RL70 -

UL intra eNB CoMP 4Rx

-

eICIC (ABS means no UL scheduling in the eICIC active cell)

Features to improve UL performance in FL16 -

Volte uplink coverage boosting

-

Configurable UL interference regions

-

64 QAM in UL

-

UL Multi cluster Scheduling

Features to improve UL performance in FL16A -

Uplink Carrier Aggregation - 2CC

-

CA-aware Idle Mode Load Balancing

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

43

Throughput Optimization Slide 45

LTE1495 – Fast Uplink Adaptation (F-ULA) Before, E-ULA (LTE1034) as open loop link adaptation ACK/NACK upper

FUG event

Increase MCS +1

EDG event

Decrease MCS -1

(Fast upgrade)

ΔC lower threshold is reached

The mechanism to trigger a FUG or an EDG event is the same in E-ULA and FULA and is based on the defined BLER target.

(Emergency downgrade)

After, F-ULA (LTE1495)

SINR is calculated based on measurements

Final MCS

SINR to MCS lookup table

+ OLLACF

FUG event

Increase OLLACF +1

(Fast upgrade)

SRS

DM-RS

ΔC

When threshold is hit

EDG event (Emergency downgrade)

Decrease OLLACF -1

(+/- ATBCF )

ATB is influenced by the power limitation of the UE

ACK/NACK

45

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

44

Throughput Optimization Slide 46

LTE1495 – Fast Uplink Adaptation (F-ULA) System level simulation results

FULA immediate MCS adaptation

FULA high UE TP available

EULA slow MCS increasing

EULA lower UE TP during transition phase

1 UE attached in good radio conditions

46

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

45

Throughput Optimization Slide 47

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

47

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

46

Throughput Optimization Slide 48

DL APN-AMBR Limits DL Throughput • Example, measurement with Samsung, 20MHz, DX HLR M14.6 • APN-AMBR DL = 20Mbps in HSS, SAE-GW APN-AMBR shaping on, FTP download # fsclish -c "show config fsClusterId=ClusterRoot fsFragmentId=FlexiNG fsFragmentId=Internal" |grep Qos GPRS AND LTE SUBSCRIBER DATA HANDLING COMMAND <

fngQosShaping: enable fngQosPolicy: enable

MAIN LEVEL COMMAND < MNI:IMSI=244070109100221; LOADING PROGRAM VERSION 7.9-2

PDN CONTEXT PARAMETERS

COMMAND EXECUTED

244070109100221 10 BOTH test-ap-2 6 2 DISABLED DISABLED YES DYNAMIC NORM 20000 65535

PHY tput, CINR 30

25

PHY tput Megabits/sec, CINR dB

IMSI ........................ CTXID ....................... PDN TYPE .................... AP NAME ..................... QOS CLASS IDENTIFIER ........ PRIORITY LEVEL .............. PRE-EMPTION CAPABILITY ...... PRE-EMPTION VULNERABILITY ... VPLMN DYNAMIC ADDRE ALLOWED . PGW ALLOCATION TYPE ......... CHARGING CHARACTERISTICS .... AMBR DOWNLINK ............... AMBR UPLINK .................

20Mbps limit on tput

20

Data Average of Phy DL TP(Mbps) Sum of SCell-CINR(Com)

15

10

5

HSS APN-AMBR limitation 06/11/2010 11:31:37.393

0 06/11/2010 11:31:47.395

06/11/2010 11:31:57.894

06/11/2010 11:32:07.895

Time

48

RA47054EN16AGLA0

© Nokia 2016

NG sw in the test: FSPR5CAT_1.39.1.1_26_r40018

RA47054EN16AGLA0

47

Throughput Optimization Slide 49

UL APN-AMBR Limits UL Throughput •

Latest FlexiNS sw implements signaling of UL APN-AMBR from MME to UE.

•

APN-AMBR parameter stored in HSS default APN profile

•

When UE receives this IE in bearer setup, it shapes the UL bit rate to the UL APNAMBR value for non-GBR bearers (per-APN)

•

Strong degrading impact on UL FTP throughput possible because UE shaping implementation is not specified in detail by 3GPP

•

Problem symptom noticed in a project: UL tput collapses after sw upgrade in MME… -

49

Fix: change APN-AMBR parameters in HSS

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

48

Throughput Optimization Slide 50

UL APN-AMBR Limits UL Throughput, Example •

Example UL APN-AMBR shaping

UL throughput goes on-off due to UE UL APN-AMBR shaping

50

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

49

Throughput Optimization Slide 51

UE-AMBR Limits •

UE aggregated maximum bits rate (AMBR) is controlled by eNB.

•

UE-AMBR is signaled to eNB in S1AP Initial Context Setup.

•

In the case of HO, it is signaled to the target cell in Handover Request.

UE-AMBR

51

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

50

Throughput Optimization Slide 52

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP throughput measurement to check that radio and e2e network do not limit peak throughput - How to troubleshoot packet loss related TCP tput problems?

52

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

51

Throughput Optimization Slide 53

Troubleshooting Checklist – Executive Summary! 1.

Check with UDP download If UDP tput not ok  problem in radio parameters or maybe interference. Use scanner to check that there are no neighbors within >30dB of serving cell. Do not trust RS CINR measurement!! Serving cell RSRP should be >-80dBm for peak tput. Check HSS profile that there is no APN-AMBR or UE-AMBR limitation • If UDP tput ok  problem is either i) packet loss in transport network, or ii) TCP settings of FTP server and/or UE. Change to Linux FTP server and optimize UE TCP settings. Check that you have no packet loss in transport network. •

53

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

52

Throughput Optimization Slide 54

UDP Testing – Motivation • in >50Mbps regime FTP throughput suffers greatly from: • Packet loss, even very little packet loss!! • Too small TCP RX and TX window

- BTS parameters can also have misconfiguration - If FTP throughput is bad, it is often difficult to say if this is caused by radio, transport or TCP setting problem - Solution: test with UDP! - If UDP throughput is also bad  then it’s most probably a radio problem • Either interference or parameters are wrong

• Could also be a problem with HSS profile - If UDP throughput is good  then it’s transport or TCP problem • Troubleshooting can be narrowed down

54

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

53

Throughput Optimization Slide 55

UDP versus TCP, Same Drive Route For Both •

74Mbps average

50Mbps average

If you must use FTP, then having multiple parallel FTP sessions may help. 55

RA47054EN16AGLA0

© Nokia 2016

UDP average = 74Mbps TCP average = 50Mbps (single FTP session)

RA47054EN16AGLA0

54

Throughput Optimization Slide 56

Please don’t forget to check radio interface DL buffer utilization: Qualcomm measurement example

UE only scheduled about 600-700 subframes a second, non-full eNB transmit buffer!! Problem with spurious TCP tx timeouts and/or packet loss??

56

RA47054EN16AGLA0

© Nokia 2016

Need MAC TTI tracer or LG for this, Samsung does not provide TTI-level data.

RA47054EN16AGLA0

55

Throughput Optimization Slide 57

Index Throughput optimization - KPIs - Downlink • Impact of inter-cell interference on peak throughput • Improving MIMO performance • Packet drops in transport/core network • Other DL issues - Uplink • User throughput vs. cell throughput • UL power control • UL scheduling • Other UL issues - QoS parameters limiting throughput Throughput testing - UDP tput measurement to check that radio and e2e network do not limit peak tput - How to troubleshoot packet loss related TCP tput problems?

57

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

56

Throughput Optimization Slide 58

How to check if the low FTP tput is caused by TCP? 1.

Try with iPerf UDP server download and upload –

If UDP speed is ok, then this is the benchmark against which TCP should be optimized

–

iPerf UDP server typically used

2.

58

Test with FTP transfer –

FTP should reach transfer speed of within a few percents of UDP speed if properly optimized

–

If the FTP download speed is much less than the UDP download, then there is most likely a problem with TCP settings or there is packet loss in the transport network somewhere

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

57

Throughput Optimization Slide 59

FTP tput problem - root causes - There are three typical root causes for bad FTP tput 1.

Windows FTP server is used in downloads (tx window too controlled)

2.

UE laptop RX window too small

3.

Packet loss in the e2e TCP path

- The first two problems are fixed by using Linux FTP server and optimizing laptop TCP settings - The third problem is: • How to detect if packet loss is the reason for bad FTP tput? • How to find where packet loss takes place?

• Rationale: Customer frequently needs to be convinced of the packet loss and/or there’s a need to pinpoint which trs box causes it

59

RA47054EN16AGLA0

© Nokia 2016

https://sharenet-ims.inside.nsn.com/Overview/D422813994

RA47054EN16AGLA0

58

Throughput Optimization Slide 60

TCP troubleshooting, how to find where packet loss takes place? Analysis steps 1.

2.

Take traces from several observation points simultaneously from the same FTP transfer. For example:



FTP server



EPC site router



eNB



UE



Transport network (output of that suspicious box..)

Use tcptrace (or some other tool) to analyze the tcp flow to find out where the packet loss happens 

60

Often the place of packet loss can be just guessed…

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

59

Throughput Optimization Slide 61

TCP troubleshooting, how to find where packet loss takes place? Extremely simplified example of the basic idea, UE making an FTP download: 

Trace from FTP server (next slide = same tcp flow from UE) TCP connection 2: host c:

89.204.133.90:1299

host d:

10.207.7.10:47459

complete conn: RESET

(SYNs: 2)

e2e transport cloud

(FINs: 1)

first packet:

Mon Nov 29 19:38:21.682101 2010

last packet:

Mon Nov 29 19:40:19.531557 2010

elapsed time:

0:01:57.849456

FTP server

Wireshark tracing point

total packets: 111515 filename:

101129_ftpdl_server 10.207.7.10.pcap

c->d: total packets:

resets sent:

55

total packets:

resets sent:

0

96729

ack pkts sent:

pure acks sent:

96727

pure acks sent:

sack pkts sent:

2115

sack pkts sent:

0

14731 1

dsack pkts sent:

0

dsack pkts sent:

0

max sack blks/ack:

2

max sack blks/ack:

0

unique bytes sent:

0

unique bytes sent: 260471616

actual data pkts:

0

actual data pkts:

actual data bytes:

0

actual data bytes: 260875344

rexmt data pkts:

0

rexmt data pkts:

rexmt data bytes:

0

rexmt data bytes:

300 403728

0

zwnd probe pkts:

0

zwnd probe bytes:

0

zwnd probe bytes:

0

outoforder pkts:

0

outoforder pkts:

pushed data pkts:

0

pushed data pkts: SYN/FIN pkts sent:

RA47054EN16AGLA0

300 retransmissions observed at FTP server

14729

zwnd probe pkts:

1/1

UE

14731

ack pkts sent:

SYN/FIN pkts sent:

61

d->c: 96784

eNB

0 1666 1/0

© Nokia 2016

RA47054EN16AGLA0

60

Throughput Optimization Slide 62

TCP troubleshooting, how to find where packet loss takes place? Extremely simplified example of the basic idea , UE making an FTP download : Trace from UE (prev slide = same tcp flow from FTP server)



TCP connection 2: host c:

89.204.133.90:1299

host d:

10.207.7.10:47459

complete conn: RESET first packet:

(SYNs: 2)

e2e transport cloud

(FINs: 1)

Mon Nov 29 19:38:41.701607 2010

last packet:

Mon Nov 29 19:40:39.551296 2010

elapsed time:

0:01:57.849688

FTP server

total packets: 290508 filename: total packets: resets sent:

d->c: 97580 55

total packets: resets sent:

0

97525

ack pkts sent:

pure acks sent:

97523

pure acks sent:

1

sack pkts sent:

2204

Only one retransmission noticed at UE end  packet loss happening somewhere in between the two observation points. Need to narrow down the actual point of packet loss by taking traces at multiple points (or make a good guess…)

192928

sack pkts sent:

0

dsack pkts sent:

0

dsack pkts sent:

0

max sack blks/ack:

2

max sack blks/ack:

0

unique bytes sent:

0

unique bytes sent: 262360656

actual data pkts:

0

actual data pkts:

actual data bytes:

0

actual data bytes: 262362016

rexmt data pkts:

0

rexmt data pkts:

rexmt data bytes:

0

rexmt data bytes:

zwnd probe pkts:

0

zwnd probe pkts:

zwnd probe bytes:

0

zwnd probe bytes:

outoforder pkts:

0

outoforder pkts:

0

pushed data pkts:

192926 1

1360 0 0

300 1666

SYN/FIN pkts sent:

1/1

SYN/FIN pkts sent:

1/0

req 1323 ws/ts:

Y/N

req 1323 ws/ts:

Y/N

RA47054EN16AGLA0

UE

192928

ack pkts sent:

pushed data pkts:

62

Wireshark tracing point

101129_ftpdl_ue.pcap

c->d:

eNB

© Nokia 2016

RA47054EN16AGLA0

61

Throughput Optimization Slide 63

Exercises

63

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

62

Throughput Optimization Slide 64

RF exercise with Nemo Outdoor Log file: Nemo log 1 rf exercise.1.nmf

Answer to following questions: •

Why DL throughput is below 20Mbit/s during the first FTP DL, around time ~13.50:00 ?

•

Why DL throughput gets better at ~13.50:40 ?

•

Why RSRQ jumps up ~13.50:45 ?

64

RA47054EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

63

Throughput Optimization Slide 65

NokiaEDU

RA47051EN16AGLA0

© Nokia 2016

RA47054EN16AGLA0

64