View
216
Download
3
Category
Preview:
Citation preview
A Comparison Between One-way Delays inOperating HSPA and LTE Networks
Markus Laner, Philipp Svoboda, Peter Romirer-Maierhofer,Navid Nikaein, Fabio Ricciato and Markus Rupp
Vienna University of Technology, AustriaFTW, Austria
EURECOM, France
WINMEE’12, May18, 2012, Paderborn, Germany
Intro. Measurements Results Interpretation Conclusion
Outline
1 Introduction
2 Measurements
3 Results
4 Interpretation
5 Conclusion
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 2/16
Intro. Measurements Results Interpretation Conclusion
Outline
1 Introduction
2 Measurements
3 Results
4 Interpretation
5 Conclusion
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 1/16
Intro. Measurements Results Interpretation Conclusion
Motivation
I Latency affects user experience
I Increasing importance/attention
I What to improve? → Measurements
Low Latency
Future Appl.?Video
Gaming
VoiceM2M
Alarm
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 2/16
Intro. Measurements Results Interpretation Conclusion
Contributions
I Measurements of LTE and HSPA latency
I Verification of LTE design goals (popular estimates)I Comparing influences of
− Packet size− Data rate
I QoS estimates for latencysensitive applications
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 3/16
latency estimates [Holma, 2010]
20
0
30
10
RTT (ms)
HSPA
NodeB
TTI
UE
Buffering
Retransm.
Scheduling
UE
LTE
TTI
Buffering
Retransm.
Scheduling
Core
RNC
eNodeB
HSPA High Speed Packet AccessLTE Long Term Evolution
RTT Round Trip TimeQoS Quality of Service
Intro. Measurements Results Interpretation Conclusion
Outline
1 Introduction
2 Measurements
3 Results
4 Interpretation
5 Conclusion
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 3/16
Intro. Measurements Results Interpretation Conclusion
Definition of LatencyI Latency [3GPP, TS 25.913]:
− Control-plane: Transition time from idle to connected state− User-plane: Packet (0 Bytes) available at IP layer of UE/GW
I Ambiguities:− Size: IP datagrams vs. IP fragments− Availability: Start vs. end− Intermediate interfaces: No IP
I Def.: Last bit of an IP datagram passing each interface
I Q: Is this def. enough for low data rate applications?
IP datagram, 4 fragmentsInterface 1
Interface 2
Interface 3
time
∆1,2 ∆2,3
∆1,3
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 4/16
Intro. Measurements Results Interpretation Conclusion
Setup
I 3G and 4G networks (3GPP Rel. 8)
I Modem: Huawei E392, triple-mode
I FDD: 10 MHz (HSPA) vs. 20 MHz (LTE)
I Radio: Line-of-sight, 130 m
ServerServer
Server
Probe
Internet
RNC GGSN
NodeBGn
SGi
Gi
USB
∆
∆
∆
Modem
S1−U
HSPA
LTE
owd
cnran
eNodeB
ProbeProbe
Probe
ProbeSAE GW
PC
Client
PC
Server
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 5/16
Intro. Measurements Results Interpretation Conclusion
Measurement Devices
I Passive packet sniffing (wiretaps) - five Probes
I GPS synchronized
I Protocol parsing
I Packet identification → timestamp comparison
ServerServer
Server
Probe
Internet
RNC GGSN
NodeBGn
SGi
Gi
USB
∆
∆
∆
Modem
S1−U
HSPA
LTE
owd
cnran
eNodeB
ProbeProbe
Probe
ProbeSAE GW
PC
Client
PC
Server
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 6/16
Intro. Measurements Results Interpretation Conclusion
Traffic Generation
I Data-traffic influences latency
I Independent uplink and downlink
I Random packet size and inter-arrival time[IETF, RFC 2330][Fabini, 2009][Baccelli, 2007]
I Blocked measurements → fixed data-rate
Internet
HSPA
modem
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 7/16
Intro. Measurements Results Interpretation Conclusion
Outline
1 Introduction
2 Measurements
3 Results
4 Interpretation
5 Conclusion
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 7/16
Intro. Measurements Results Interpretation Conclusion
Comparison RTT
I LTE: 36 ms median RTT
I HSPA: 42 ms median RTT
I Minor improvements of 14%
0 20 40 60 800
0.5
1
overall round−trip time (ms)
em
piric
al C
DF
LTE
HSPA
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 8/16
RTT RoundTrip Time
CDF CumulativeDistribution Function
Intro. Measurements Results Interpretation Conclusion
One-way Delay in RAN
I Downlink: LTE very fast (7 ms)
I Uplink: HSPA fast (15 ms)
I Main reasons for minor performance of LTE:Low layer signaling, scheduling, DRX, and cell configuration(e.g. SR periodicity)
0 15 30 45 600
0.5
1
one−way delay RAN (ms)
em
piric
al C
DF
LTE downlink
LTE uplink
HSPA downlink
HSPA uplink
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 9/16
RAN Radio AccessNetwork
CDF CumulativeDistribution Function
Intro. Measurements Results Interpretation Conclusion
One-way Delay in Core
I Uplink and Downlink equal (1.5 ms)
I HSPA and LTE similar (1.5 ms)
I Caution: high load in HSPA → tail
0 1 2 3 4 50
0.5
1
one−way delay CN (ms)
em
piric
al C
DF
LTE downlink
LTE uplink
HSPA downlink
HSPA uplink
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 10/16
Intro. Measurements Results Interpretation Conclusion
LTE Delay vs. Packet Size
I Scatterplot: No diagonal components⇒ no correlation
I Reason: High bandwidth & LTE scheduling
Downlink
0 10 20 30 40 500
1
2
3
4
5
overall one−way delay (ms)
data
gra
m s
ize (
kB
yte
)
0
1
Uplink
0 10 20 30 40 500
1
2
3
4
5
overall one−way delay (ms)
data
gra
m s
ize (
kB
yte
)
0
1
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 11/16
Intro. Measurements Results Interpretation Conclusion
LTE Delay vs. Packet Size
I Scatterplot: Diagonal components⇒ positive correlation
I Reason: HSPA scheduling & network load
Downlink
0 10 20 30 40 500
1
2
3
4
5
overall one−way delay (ms)
data
gra
m s
ize (
kB
yte
)
0
1
Uplink
0 10 20 30 40 500
1
2
3
4
5
overall one−way delay (ms)
data
gra
m s
ize (
kB
yte
)
0
1
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 12/16
Intro. Measurements Results Interpretation Conclusion
Delay vs. Data-rateI LTE: independent of packet-size
I HSPA: dependent of packet-size (2 curves)
I Delay decreasing with increasing ratestrong contrast to fixed networks (queuing)
256 1k 4k 16k 64k 256k 1M0
15
30
45
60
data rate (Byte/s)
me
dia
n o
ve
rall
on
e−
wa
y d
ela
y (
ms)
LTE downlink
LTE uplink
HSPA downlink
HSPA uplink
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 13/16
Intro. Measurements Results Interpretation Conclusion
Outline
1 Introduction
2 Measurements
3 Results
4 Interpretation
5 Conclusion
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 13/16
Intro. Measurements Results Interpretation Conclusion
Latency Sensitive Traffic
Online Gaming
I Data-rate: 1 kByte/s – 16 kByte/s
I Latency: Impairments at RTT > 50 ms
M2M (e.g., event-driven, real-time)
I Data-rate: < 1 kByte/s
I Latency: Impairments at RTT > 20 ms
VoIP, video call
I Data-rate: 4 kByte/s – 256 kByte/s
I Latency: Impairments at RTT > 200 ms
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 14/16
Intro. Measurements Results Interpretation Conclusion
Network Type vs. Traffic
I QoS requirements vs. network performance
I Not accounting for IP backbone and server processing delays
I 3G/4G networks not (yet) ready to host certain apps.
I Source of QoS parameters [LOLA, D3.6, 2012]
Application LTE (up/down) HSPA (up/down)
Online Gaming ( 31 / 13 ) ms ( 12 / 17 ) msM2M ( 30 / 10 ) ms ( 10 / 16 ) msVoIP ( 30 / 15 ) ms ( 35 / 16 ) ms
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 15/16
QoS Quality of ServiceM2M Machine to MachineVoIP Voice over IP
Intro. Measurements Results Interpretation Conclusion
Outline
1 Introduction
2 Measurements
3 Results
4 Interpretation
5 Conclusion
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 15/16
Intro. Measurements Results Interpretation Conclusion
Conclusions
I LTE shows improved latency over HSPA (14%)
I Decreasing latency with increasing data-rate
I HSPA uplink scheduling outperforms LTE
I New low-latency transmission techniques [LOLA project]
I Semi-persistent scheduling in LTE
I Latency sensitive apps. require improvements
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 16/16
Intro. Measurements Results Interpretation Conclusion
Thank you for your attention!
Markus Laner, Philipp Svoboda, Peter Romirer-Maierhofer,Navid Nikaein, Fabio Ricciato and Markus Rupp
Vienna University of Technology, AustriaFTW, AustriaEURECOM, France
mlaner@tuwien.ac.atwww.tc.tuwien.ac.at
M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 16/16
Ref.
ReferencesI 3GPP. TS 25.913, Requirements for Evolved UTRA and Evolved UTRAN,
http://www.3gpp.org/.
I H. Holma and A. Toskala, WCDMA for UMTS: HSPA Evolution and LTE (5thEdition). Wiley, 2010.
I M. Laner, P. Svoboda, and M. Rupp, Dissecting 3G Uplink Delay by Measuringin an Operational HSPA Network, in PAM’11, Atlanta, Georgia, 2011.
I J. Fabini, L. Wallentin, and P. Reichl, The importance of being really random:methodological aspects of IP-layer 2G and 3G network delay assessment, inICC’09, Dresden, Germany, 2009.
I V. Paxson et al. (1998) RFC 2330, Framework for IP Performance Metrics,http://www.ietf.org/rfc/rfc2330.txt.
I F. Baccelli et al., On Optimal Probing for Delay and Loss Measurement, inIMC’07, San Diego, California, 2007.
I libpcap - library for network traffic capture, http://www.tcpdump.org/.
I Darwin Project, http://userver.ftw.at/∼ricciato/darwin/.
I Endace DAG, http://www.endace.com/.
I LinuxPPS, http://wiki.enneenne.com/index.php/LinuxPPS support.
I EU FP7 LOLA Project, http://www.ict-lola.eu/.M. Laner, TUV Comparison of Delays in HSPA and LTE 2012-05-18 17/16
Recommended