Survey
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project
A Dynamic Buffer Management Scheme for End-to-end QoS
Enhancement of Multi-flow Services in HSDPA
Suleiman Y. Yerima, Khalid Al-Begain
Integrated Communications Research Centre
Faculty of Advanced Technology
University of Glamorgan
Outline
HSDPA overview
TSP queuing and BM schemes
Time-space priority BM
2
Dynamic Time-space priority BM
Simulation model
Results
Conclusions
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
HSDPA Overview
High Speed Downlink Packet Access (HSDPA):
3GPP Enhancements to UMTS (3G) RAN
Higher Peak data rate: up to 14Mbps
Lower connection and response times
3-5 X capacity increase
External
Network
Core Network
User Equipments
Three interacting domains:
Core Network
Radio Access Network (RAN)
User Equipment (UE)
Iub interface
Node B
HSDPA CELL
3
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Radio Network
Controller
HSDPA Overview
New PHY & MAC enhancements in Node B:
New MAC-hs layer in Node B
High-Speed downlink Shared Channel
Link adaptation (AMC)
Packet Scheduling (PS)
External
Network
Core Network
User Equipments
L1 retransmissions (HARQ)
Shorter Transmission interval (2ms)
Iub interface
Node B
HSDPA CELL
4
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Radio Network
Controller
motivation:
Emergence of multiple flow traffic profile per user/connection
Existing HSDPA QoS mechanisms single flow based
Node B buffering
5
E.g. MAC-hs Packet Scheduling (PS)
No BM schemes in standards- open issue
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
motivation (contd.)
The most challenging multiple flow scenario is connections with RT and NRT
flows (conflicting QoS requirements) e.g. Voice + file download
RT flow -> delay, jitter sensitive & loss tolerance
NRT flow-> loss sensitivity & delay, jitter tolerance
Hence our proposed MAC-hs BM schemes based on Time-Space Priority
(TSP) queuing model
6
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
TSP queuing model
Typical priority queuing models are either loss
or delay differentiated
Our Time-space priority queuing model (TSP):
Single queue with hybrid differentiation
Loss differentiated
delay differentiated
RTC flow
RTC packets:
High priority delay
Low priority loss
Service
process
NRTC
flow
NRTC packets
High priority loss
low priority delay
Hence RTC => preferential transmission
NRTC =>preferential buffer admission
7
TSP
threshold
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Transmission
to user
terminal
TSP advantages
Efficient buffer utilization
Most viable for joint RTC and NRTC QoS control compared to
typical priority queuing approaches
Hence we designed an efficient TSP-based BM scheme for HSDPA
8
Exploiting existing mechanisms in HSDPA specs.
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
TSP-based Buffer Mgt. in HSDPA
Enhanced TSP with flow control thresholds
RNC sends MAC PDUs over Iub interface
Employs Iub signalling with credit allocation
UE1
N
H
Iub flow
control
algorithm to mitigate buffer overflow
UE1 MAC-hs buffer
RNC
L
UE2
Employs Discard Timer for RTC
R
Packet
Scheduling
Higher layer protocol (ARQ, TCP)
UE1 RT flow
performance improvement
UE1 NRT flow
TCP Source
UE3
UE
CN & EXTERNAL IP
Node B
RNC
Node B
Capacity Request {Priority, User Buffer Size (UBS)}
Capacity Allocation {Priority, Credits}
RNC
HARQ
HS-DSCH Frame {Priority, UBS, PDU size, #PDUs}
ARQ
Air interface
TCP
Iub interface
9
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
TSP-based BM in HSDPA
Credit based FC algorithm:
CTotal = CNRT + CRT
CRT = (λRT / PDU_size) ∙ TTI
CNRT = min { CNRTmax , RNCNRT }
CNRTmax = (λ’NRT /PDU_size) ∙ TTI ,
N’T < L
β ∙ (λ’NRT /PDU_size) ∙ TTI ,
L ≤ N’T ≤ H,
0 ,
N’T > H
0<β<1
Where λ'NRT = α ∙ λ'NRT-1 + (1- α) ∙ λNRT is EWMA of Scheduler NRT data rate
and N’T = θ ∙ NT-1 + θ ∙ NT
10
is an EWMA of total queue size
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Enhanced TSP improves e2e NRTC throughput without compromising
RT QoS
Problem: TSP has static delay prioritization
Potential NRTC bandwidth starvation
Non optimal ARQ and TCP performance
A possible solution:
Exploit possible RTC delay tolerance
Hence we extend the TSP BM with Dynamic Time priority switching
11
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
D-TSP
Incorporates delay Priority switching to TSP
RNC
UE1 MAC-hs buffer
UE1
DTSP priority switching algorithm:
N
IF RT packets < k AND RT HOL delay<
MAX_delay AND NRT packets > 0
Time Priority = NRT flow
Generate Transport Block from NRT PDUs
ELSE
Time Priority = RT flow
Generate Transport Block from RT PDUs
MAX_delay = Max e2e delay – other queuing and
propagation delays
H
Iub flow
control
L
RT
R
NRT
Priority
switching
k= Delay budget / RTC inter-arrival time
Delay budget ≤ MAX_delay
Discard timer (DT) setting = MAX_delay
12
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Packet
Scheduling
HSPDA modelling
Detailed custom modelling with OPNET:
Multi-flow connection: VoIP source, NRT source with TCP
Fixed external and Core Network delay assumed
RNC: Packet segmentation, RLC modes
Node B: AMC Link adaptation, HARQ, MAC-hs buffers, Packet scheduler
Receiver: SINR, HARQ, RLC modes, re-assembly queues, TCP
User Equipments
voice + data
connection
HSDPA
CELL
Node B
2ms
13
RNC
20ms
Core Network
External Network
70ms
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
simulation set up
HSDPA Simulation Parameters
Concurrent VoIP + FTP for 180s
• MAX_delay = 250 –( 70 – 20) = 160ms
HS-DSCH TTI
2ms
Path loss Model
148 + 40 log (R) dB
Shadow fading
Log-normal: σ = 8 dB
Number of HSDSCH
codes
5
CQI delay
3 TTIs (6ms)
HARQ processes
4
Performance metrics in test UE:
HARQ feedback delay
5ms
• End-to-end NRTC throughput
Test UE position from
Node B
0.2 km
• Voice PDU discard ratio (Discard timer)
Packet Scheduling
Round Robin
• % HSDPA channel utilization
RLC PDU size
320 bits
RNC-Node B delay
20ms
External + CN delays
70ms
TCP config
MSS= 536 bytes, RWIND = 64
Flow control settings
β = 0.5, α = 0.7, θ = 0.7
• BM config: R = 10, L =100, H= 150,
N= 200 PDUs
• DTSP params: k = 2, 4, 6, 8
• channel load: 1, 5, 10, 20, 30, 50 users
RNC
Node B
14
2ms
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
20ms
70ms
Results: UE1 Data download throughput (1 user)
220000
200000
180000
160000
Throughput at UE1
Throughput (bps)
140000
• DTSP and TSP show similar
performance
120000
• Increased DB settings has
marginal effect on throughput
100000
80000
• very low HS-DSCH load hence no
DTSP gain
60000
40000
20000
0
0
18
36
54
72
90
108
126
144
162
180
Time (s)
TSP
15
D-TSP (k = 2)
D-TSP (k = 4)
D-TSP (k = 6)
D-TSP (k = 8)
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Results: UE1 Data download throughput (5 users)
160000
150000
140000
130000
120000
Throughput at UE1
Throughput (bps)
110000
100000
• More load on HSDPA channel
90000
• Increased DB settings show
noticeable improvement
80000
70000
• DTSP performs better than TSP
60000
50000
40000
30000
20000
10000
0
0
18
54
36
72
90
108
126
144
162
180
Time (s)
TSP
16
D-TSP (k = 2)
D-TSP (k = 4)
D-TSP (k = 6)
D-TSP (k = 8)
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Results: UE1 Data download throughput (10 users)
120000
110000
100000
90000
Throughput at UE1
Throughput (bps)
80000
• Higher load on HSDPA channel
70000
50000
• Increased DB settings show
noticeable improvement
40000
• DTSP performs better than TSP
60000
30000
20000
10000
0
0
18
36
54
72
90
108
126
144
162
180
Time (s)
TSP
17
D-TSP (k = 2)
D-TSP (k =4)
D-TSP (k = 6)
D-TSP (k = 8)
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Results: UE1 Data download throughput (20 users)
80000
75000
70000
65000
60000
Throughput at UE1
Throughput (bps)
55000
50000
• Higher load on HSDPA channel
45000
• Increased DB settings show
noticeable improvement
40000
35000
30000
• DTSP performs better than TSP
25000
20000
15000
10000
5000
0
0
18
36
54
72
90
108
126
144
162
180
Time (s)
TSP
18
D-TSP (k =2)
D-TSP (k = 4)
D-TSP (k = 6)
D-TSP (k = 8)
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Results: UE1 Data download throughput (30 users)
55000
50000
45000
40000
Throughput (bps)
35000
Throughput at UE1
30000
• Higher load on HSDPA channel
25000
• Increased DB settings show
noticeable improvement
20000
15000
• DTSP performs better than TSP
10000
5000
0
0
18
36
54
72
90
108
126
144
162
180
Time (s)
TSP
19
D-TSP (k = 2)
D-TSP (k = 4)
D-TSP (k =6)
D-TSP (k = 8)
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Results: UE1 Data download throughput (50 users)
34000
32000
30000
28000
26000
24000
Throughput at UE1
Throughput (bps)
22000
20000
• Higher load on HSDPA channel
18000
• Increased DB settings show
noticeable improvement
16000
14000
12000
• DTSP performs better than TSP
10000
8000
• performance peaks at k = 6
6000
4000
2000
0
0
18
54
36
72
90
108
126
144
162
180
Time (s)
TSP
20
D-TSP ( k =2)
D-TSP (k =4)
D-TSP (k = 6)
D-TSP (k = 8)
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Results: Voice pdu discard ratio vs DB settings
0.28
0.26
Voice pdu discard loss
0.24
• assuming max DR= 2%
% of VoIP PDUs dropped by Discard Timer
0.22
0.18
•In 1, 5, and 10 user
scenarios VoIP QoS satisfied
0.16
•Optimum k for 20 users = 6
0.14
• optimum k for 30 users = 4
0.12
•Optimum k for 50 users = 2
0.2
0.1
0.08
0.06
0.04
0.02
0
TSP
D-TSP (k=2)
1 user
21
5 users
D-TSP (k=4)
10 users
20 users
D-TSP (k=6)
30 users
D-TSP (k=8)
50 users
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Results: HSDPA channel utilization vs DB settings
0.68
0.66
UE 1 HSDPA Channel Utilization
0.64
UE1 HSDPA channel
utilization
0.62
0.6
•Utilization constant in DTSP
regardless of DB setting in 1
user scenario
0.58
0.56
• channel utilization
improves with higher load
due to pdu bundling
0.54
0.52
0.5
0.48
TSP
D-TSP (k =2)
1 user
22
5 users
D-TSP (k=4)
10 users
20 users
D-TSP (k=6)
30 users
D-TSP (k=8)
• DTSP has better channel
utilization than TSP except at
very low load ( e.g. 1 user
scenario)
50 users
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Conclusions and Summary
Conclusions:
DTSP achieves e-2-e throughput improvement for the multi-flow
NRTC traffic
Better channel utilization is achieved with DTS P over TSP
Acceptable VoIP performance within QoS constraints
Further work
Investigate with other Packet Scheduling
Investigate other possible multi-flow scenarios
23
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008
Thank You!!!
24
NGMAST ’08 International Conference, University of Glamorgan, Cardiff, Wales, Sept19th 2008