Transcript Estimating end-to-end performance in 3G Long

Estimating end-to-end performance in 3G
Long-Term Evolution compared to HSDPA
Thesis work seminar presentation 18.10.2005
Mari-Jaana Pelkonen 51529B
Company Confidential
Acknowledgement
• Supervisor: Prof. Heikki Hämmäinen
• Instructor: Jani Kokkonen M.Sc
• Nokia Networks, System Technologies
Company Confidential
Agenda
• Thesis introduction
• HSDPA overview
• 3G LTE overview
• Estimation work
• Summary
Company Confidential
Thesis introduction
• 3G Long-Term Evolution standardization effort started in late 2004 in 3GPP
• 3G networks are implemented at very slow phase. One major reason for the
operators low investment willingness is the low capacity it offers to the operator and to
the customer.
• IEEE is standardizing mobile WiMAX => Threat for loosing competitive edge.
• In Japan the telecom technology is one step forward: DoCoMo is driving the
standardization.
• Why not 4G? 4G will be a system that connects all the existing and future networks
seamlessly together. The technology is not yet ready for that. 3G LTE is a evolution
step towards the 4G, enabling the operators to use the existing infrastructure longer.
• Target to standardize simple, IP optimize network, offering mobile DSL type
connections.
Company Confidential
Thesis introduction
• The scope of the thesis work was to prove that the performance in presented 3G
LTE architecture is better than in the current available systems.
• 3G HSDPA was selected to the reference architecture.
• We were not only interested whether the new system is better, but why and why
not.
• How much of the improvement could be achieved only improving capacity of the
legacy systems?
• What is the impact of the new architecture solutions
• Different applications have different requirements for the network, performance is
application specific. Therefore delay and throughput impact estimations were
done for three applications: Web browsing, streaming video and VoIP.
Company Confidential
Thesis introduction
• This thesis work was written in 3G Long-Term Evolution architecture project
• One part was literature study about 3G HSDPA performance and performance in
general.
• 3G LTE specific part is taken from the architecture project and standardization
contributions. The 3G LTE architecture presented in this work is DRAFT
architecture. It will not be standardized as presented here.
• The estimation work is done using a Service performance Excel tool created to
calculate delays in 3G networks. The tool consists of signaling flows for different
applications. For that work, the 3G LTE specific parts were added to the tool.
• Values used in the tool are for 3G networks measured or estimated. To get 3G
LTE values, I consulted several experts working with that area. Some of the
values are targets, other derived from 3G values and the rest are educated
guesses.
Company Confidential
HSDPA
• High-Speed Downlink Packet Access is 3G performance enhancement
technology. It does not change the core network, but only the radio interface in
the downlink direction.
• HSDPA offers theoretical DL bit rates up to 14.4 Mbps.
• Only test networks implemented, not yet in commercial use. The effective bit rate
offered to users is assumed to be around 800 kbps.
Company Confidential
HSDPA: 3G architecture
Uu
UE = User
Equipment
UTRAN Iu
Node B = base
station
CN
Registers
RNC = Radio
Network Controller
HLR AuC EIR
RNS = Radio
Network System
RNS
Node
B
Iub
CN = Core Network
UE
RNC
UTRAN = Universal
Terrestrial Radio
Access Network
CN PS domain
Node
B
Iub
SGSN
GGSN
G
n
Gi
SGSN = Service
Gateway Supporting
Node
GGSN = Gateway
GGSN Supporting
Node
Company Confidential
HSDPA: 3G QoS bearer architecture
MT
TE
UTRAN
CN Iu edge
CN Gateway
TE
End-to-end Service
Local Bearer
Service
Radio Access Bearer Service
Company Confidential
External Bearer
Service
UMTS Bearer Service
Radio Bearer
Service
Iu Bearer
Service
UTRA Service
Physical Bearer
Service
CN Bearer
Service
Backbone
Bearer Service
HSDPA: Protocol stack (user plane)
HLR
UE
U
BS
Iub
RNC
3G-SGSN
IuPs
Gn
GGSN
Server
Application
Application
TCP/UDPu
TCP/UDP
IPv6/v4
IPv6/v4
PDC GTPRLCP
U
MAC- UDP
U
PDC
RLCP
U
MAC
Radio
L1
Company Confidential
Gr
HSDSCH
MAC-hs FP
Radio
L1
GTP- GTPU
U
UDP UDP
GTPU
UDP
IPv6/v4
L2
L2
L2
L1
L1
L1
DHSDSCH
FP
IP
IP
IP
IP
L2
L2
L2
L2
L2
L2
L1
L1
L1
L1
L1
L1
HSDPA: WCDMA RRC and PMM states
RRC state change
CELL_PCH
PMM
Connected
GPRS
Attach
If DL activated,
paging causes
delay
DCH channel
allocation time
CELL_DCH
CELL_FACH
2-5 s timer
2-5 s timer
RRC Connection establishment time
PMM Detached
IDLE
Mobile is allowed to send
data in CELL_FACH and
CELL_DCH states. DCH
channel is dedicated
channel for end user
data.
CELL_PCH and
URA_PCH (not shown in
the figure) are used for
paging.
In idle mode mobile has
no radio connection.
Company Confidential
3G LTE
• IP optimized network architecture
• Target is to solve the performance problems that current 3G architecture has and
offer DSL type mobile internet connection.
• Simple architecture
• Short user plane RTT
• Cell capacity up to 100 Mbps
• In between 3G and 4G, interworking with existing and future network
technologies inbuilt in the architecture.
Company Confidential
3G LTE: Goodbye circuit switched voice!
• The evolution of packet switched network technology has made possible to
transmit voice over IP network with acceptable end-user performance.
• The SKYPE is one of the most popular example of that.
• Current 3G and 2G networks are optimized for circuit switched voice, that makes
them complex and not best possible for data traffic.
• Operators need to invest in and maintain two parallel networks: CS and PS.
• The all-IP architecture will be simple and cheap!
• Of course operators are not willing to cannibalize their CS voice business by
offering VoIP. The success of SKYPE shows, that former or later customers are
changing to the VoIP. To ensure not to loose the future profit, operators need to
be inside the VoIP business.
Company Confidential
3G LTE reference architecture
UE = User Equipment
BS = Base Station
RNC
functionalities
moved in the
base station.
SN-C = Serving Node
(Control plane)
Internet
Subscription
AAA Registers
Serving
Node - C
BS
Access Network
Serving
Node - U
Inter -connection
Service
Gateway
HA
BS
Company Confidential
Operator
service
network
SN-U = Serving Node (User
plane)
SGW = Service Gateway
3G LTE: QoS bearer architecture
UE
SN
BS
User- IP
Tunneling or forwarding
Radio
Transport
Transport
Note: this is called bearerless compared to current 3G bearer
architecture. Air interface connection establishment and
modification is simplified by reducing the number of air-interface
bearers.
Instead of four radio bearers, only one radio bearer has to be
established. This leads to the significantly reduced radio
connection setup time.
Company Confidential
3G LTE: Mobility Management States
Connection
Failure, UE
local release
AssignUE_LLA,
AssociateRLID
Release
Release
RLID
Idle
Active
Associate
RLID
Detached
Release UE_LLA &
RLID
 The number of channels reduced. Only one channel for user data. That
channel is associated, if UE is in Active state.
 That allows to reduce the number of states to three. If user is connected
to the network, it is Idle or Active, whether it has data to send or receive.
Company Confidential
3G LTE: Protocol stack (User plane)
All-IP protocol architecture, one continuous IP layer through all
the network elements.
Company Confidential
Estimation work
• Throughputs, link utilizations and transfer delays for TCP is estimated for different
file sizes.
• Studied applications were VoIP, web browsing and streaming.
• For VoIP call, the most critical Key Performance Identifiers are session setup
delay, end-to-end delay and delay variation. Session setup delay and end-to-end
delay were estimated.
• For web browsing, the KPI studied is the click-to-content time, i.e. the time that
takes after user selects page until it is loaded to his computer.
• KPIs for streaming are session setup delay and the throughput. Because
throughput is studied separately, only session setup delay is estimated.
Company Confidential
Estimation work: TCP throughputs
 TCP throughput for
3G LTE (800 kbps) is
better with all file sizes
than HSDPA
Average throughput in HSDPA and 3G LTE
Average TCP throughput (kbps)
7000
6000
5000
HSDPA (800)
4000
3G LTE (6000)
3G LTE (3000)
3000
3G LTE (1500)
3G LTE (800)
2000
1000
0
12
200
Size of file (kB)
Company Confidential
1000
 Due the TCP slow
start effect, the TCP
throughput is worse
with small files than the
large ones.
Estimation work: TCP Link utilizations and delays
TCP delay in HSDPA and 3G LTE
Link Utilization in HSDPA and 3G LTE
14
120
12
HSDPA (800)
80
3G LTE (6000)
60
3G LTE (3000)
3G LTE (1500)
40
3G LTE (800)
20
TCP delay (s)
Link utilization (%)
100
10
HSDPA (800)
8
3G LTE (6000)
3G LTE (3000)
6
3G LTE (1500)
3G LTE (800)
4
2
0
0
12
200
Size of file (kB)
1000
12
200
Size of file (kB)
3G LTE link utilization with same bit rate is notable better.
Company Confidential
1000
Estimation work: Streaming session setup time
Phase
3G LTE delay (ms)
HSDPA with always on
PDP context (ms)
HSDPA delay (ms)
RTSP signaling
329
535
535
TCP connection establishment
77
156
156
Primary PDP context without RAB
-
-
769
1408
1408
RAB establishment
Secondary PDP context with RAB
-
1975
1974
Delay before buffering
406
4073
4843
Buffering
5000
5000
5000
Total
5406
9073
9843
Company Confidential
Estimation work: VoIP call with Internet Multimedia
Subsystem
System
SIP session setup delay
(ms)
End-to-end delay
(ms) (for 210
bytes VoIP packet)
RTT UE1-UE2-UE1
(ms)
3G LTE
2385
34
68
HSDPA with
always on PDP
context
7894
97
194
Difference
5509
63
126
 For 3G LTE the SIP session setup delay is less than the circuit
switch PSTN call setup delay.
-External network delay
not calculated
- Both end-users are
connected to their own
IMSs.
 The difference ín session setup delay is 5.5 second. Most of the
difference is caused by the secondary PDP context activation and
RAB procedures.
 End-to-end delay for 3G LTE 30 ms is not notable for user. HSDPA
71 ms end-to-end delay is not notable with echo cancellation.
Company Confidential
Estimation work: VoIP End-toEnd Delay
3G-SGSN
3G
RNC
UE
Node B
IMS 1
3G-SGSN
RNC
UE
IP/MPLS/IPoATMbackbone
GGSN
Node B
IP/MPLS/IPoATMbackbone
GGSN
IMS 2
 End-to-end delay consists of processing delays in UEs and in every
network node in between them and transition delays between nodes.
Company Confidential
Estimation work: Web browsing
First Page Delay
Second page delay
12000
12000
HSDPA (800/128)
10000
HSDPA with always on
PDP context
8000
3G LTE (6000/2500)
6000
3G LTE (3000/1500)
4000
3G LTE (1500/512)
Total delay (ms)
Total delay (ms)
10000
HSDPA (800/125)
3G LTE (6000/2500)
8000
3G Lte (3000/1500)
6000
3G LTE (1500/512)
4000
3G LTE (800/368)
3G LTE (512/2576)
2000
2000
3G LTE (800/384)
0
0
40
300
600
Size of page and objects (kB)
3G LTE (512/2576)
40
300
Size of page and objects (kB)
 Estimation is done for HSDPA with and without always-on
PDP context.
 First page delay includes radio connection establishment,
PDP context activation and DNS query
Second page delay consists only HTTP signaling.
Company Confidential
600
First page delay divided into parts
First page delay, page and objects total 40 kB
4000
3500
time (ms)
3000
DNS Query
2500
RAB/Radio connection
2000
PDP context
1500
HTTP protocol
1000
500
0
HSDPA HSDPA 3G LTE 3G LTE 3G LTE 3G LTE 3G LTE
(800)
with
(6000) (3000) (1500)
(800)
(512)
always
on PDP
context
Company Confidential
Summary
• Performance advantage of presented 3G LTE is clear for investigated
applications.
• The session setup delay (PDP context and radio connection establishment) in 3G
affects worst in short living applications, or applications that transfers only small
amount of data.
• Enhanced air- interface effect is notable only with applications that transmit large
files
• The capacity increase or RTT decrease is not the only way to the better
performance. The IP connectivity added with bearerless model presented here is
needed to reduce the session setup latencies.
Company Confidential