Transcript Remi Ando, Tutomu Murase, Masato Oguchi

Characteristics of QoS-Guaranteed TCP on
Real Mobile Terminal in Wireless LAN
Remi Ando†
Tutomu Murase‡
Masato Oguchi†
†Ochanomizu
University,Japan
‡ NEC Corporation
Outline

Background



Characteristics of AP




Buffer sizes of APs
Previous Researches about QoS-TCP
Problem and experiment on QoS control
Experimental Result




QoS-TCP
Unfairness problem of TCP throughput on Wireless LAN
Experiment on a real environment
Comparison with computer simulation
Analysis of performance evaluation results
Conclusion
2011/05/12
CQR2011
2
Background
IP
IP
network
MAC
WLAN
application
QoS pass
Streaming server


Spread of Wireless LAN
An increased demand for multimedia communications
Various controls
proposed in each
protocol layer
⇒QoS control in wireless LAN environment becomes important
2011/05/12
CQR2011
3
Background
IP
IP
network
MAC
WLAN
application
QoS pass
Streaming server
×Application：cannot correspond to many or new applications
×IP：necessary to change in the entire the network
×MAC：change of wireless LAN equipment
◎TCP：control on the network edge
QoS-TCP(tries to guarantee the
bandwidth based on TCP)
firewall can be passed
2011/05/12
CQR2011
4
QoS-TCP


Aiming at quality improvements of streaming
communications
Designed to assure a designated bandwidth



ssthresh is flexibly set by using the target bandwidth
cwnd is led to keep the target bandwidth, and kept large as
much as possible even in packet loss
it avoids congestion collapsing
cwnd
Packet loss
Packet loss
QoS-TCP
TCP
ssthresh
2011/05/12
CQR2011
5
TCP throughput Oligopoly
(Unfairness problem of TCP throughput)
TCP throughput is not evenly divided among terminals


Higher throughput flow can send more, and Lower
throughput flow can send less
The number of the terminals that becomes fair or unfair
changes by buffer size of AP
7
unluckey
terminal
10
Throughput(Mbps)

No data losses and TCP ACK overflow at AP buffer causes this unfairness
8
Thfoughput(Mbps)

6
4
2
0
1
2
Terminal ID
3
6
5
4
3
2
1
0
1
fair
2011/05/12
2
3
4
5
Terminal ID
6
7
unfair
CQR2011
6
Characteristics of AP

Evaluation environment

sender

wired<100M>
Buffer sizes of APs affect
much on TCP/QoS-TCP
throughput characteristics
Difficult to know buffer
sizes of off-the-shelf APs

AP
wireless<IEEE802.11g>

Estimated the buffer sizes

receiver
2011/05/12
Vendors do not disclose
such information
CQR2011
Imposing excess traffic to
the AP and comparing
input packets with output
packets of the APs
7
Buffer sizes of APs
Result of a measurement
 265.5 packets

Planex CQW-MR500



256 packets


135.1 packets



BUFFALO WHR-HP-AMPG
90.5 packets


BUFFALO WZR-AMPG300NH
We should carefully
decide buffer size in
simulation
NEC PA-WR8500N
37.2 packets

The buffer size is an
average of 10 times
Buffer sizes of the 5 APs
vary between about 30
and 300 packets
Buffer size effects on TCP
throughput characteristics
BUFFALO WHR-AM54G54
2011/05/12
CQR2011
8
Previous Researches about QoS-TCP

Simulation and Real environments
effectiveness
has already
been verified
wired networks
 wireless networks(fixed)
→An evaluation on moving terminals has not been performed yet


Evaluation of wireless networks(moving)
Measurements of normal TCP throughput on mobile teminals
have already done
→ characteristics of QoS-TCP has not known yet

Characteristics of QoS-TCP on mobile
terminal in wireless LAN
2011/05/12
CQR2011
9
Problem and experiment on QoS control

１．Moving of mobile terminals influences the bandwidth
guarantee of QoS-TCP

Radio field strength changes


The number of the terminals changes


disadvantage in the radio wave compared with the terminal that is
already communicating
fair↔unfair
２．Quality degration is confirmed to make QoS-TCP minimum

quality degradation happens to in the case of switching AP at Handover

Comparison of TCP and QoS-TCP
Terminals in an environment
of indoor/outdoor and
simulation
2011/05/12
CQR2011
10
Higher throughput flow can
send more, and
Lower throughput flow can
send less
Evaluation model
QoS-TCP
handover
20m
AP1
<802.11g>
Each data is sent in uplink TCP
flow direction
AP2
<802.11g>
<100M>
2 data sending
terminals
6 data sending
terminals
server
6 terminals in total do not cause unfair situation and
7 terminals in total do cause unfair situation
2011/05/12
CQR2011
11
Channel capacity measurement

Radio interferences are measured in TCP maximum throughput
One mobile terminal is used to measure

Outdoor


25
30
25
15
TCP-up
10
Throughput(Mbps)
20
Thfoughput(Mbps)
Indoor
5
20
15
TCP-up
10
5
0
0
AP1
AP2
AP1
AP2
100m
20m

The interferences is strong enough
to reduce the link capacity even it
is close to the APs
2011/05/12
The emission power of the radio
signal is adjusted in order to be
well attenuated at the point of the
CQR2011HO
12

Throughput QoS-TCP and TCP (outdoor)
Throughput of mobile node(Mbps)
16
14
12
10
Hand
Over
8
6
QoS-TCP case
TCP case
4
2
0
0
5
AP1



10 15 20 25 30 35 40 45 50 55 60 65 70 75 80
AP2
Distance from mobile terminal to AP(m)
TCP:Throughput is always less than or equal to the fair-share value
QoS-TCP :slightly more than fair-share throughput near AP1 , but
throughput fails near AP2 and anywhere else
The radio interference dominates to decide , so QoS-TCP fails to
guarantee a bandwidth in this case
2011/05/12
CQR2011
13
Throughput QoS-TCP and TCP (indoor)
12
Throughput(Mbps)
10
QoS-TCP gets
bandwidth
8
Hand
Over
6
QoS-TCP
TCP
4
2
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Distance from mobile terminal to AP(m)
3 competitive
terminals
7 competitive
terminals
TCP cannot acquire bandwidth after the handover, but
- QoS-TCP can
⇒ Even QoS-TCP joins the competition after 6 terminals dominate
whole bandwidth, QoS-TCP can defeat the oligopoly.
-
2011/05/12
CQR2011
14
Comparison with computer simulation
Throughput(Mbps)
12
Difference about
12 times total
throughput
10
8
6
simulation
4
real terminal
HO
2
0
1 2 3 4 5 6 7 8 9 1011121314151617181920
HO
Distance from mobile terminal to AP(m)

About 12 times difference even though similar
parameters are used
2011/05/12
CQR2011
15
Analysis of performance evaluation results(1/2)

sender

Congestion window
Packet dump data


QoS-TCP
Air
Wired
TCP-Proxy
converter converter
converter
converter
<wireless LAN>
（IEEE802.11g）
AP
<wired>100M
receiver
2011/05/10
CQR2011
16
Analysis of performance evaluation results


Very rare but bursty MAC data loss event occurred in the air
No throughput degradation was aware (but really happened)
throughput
throughput
throughput
seconds

mili seconds
minutes
Good chance to grab bandwidth for new comers!!
 Larger slow start threshold in QoS-TCP than TCP
 So, QoS-TCP can rump up faster than TCP and defeat TCP
Simulation: QoS-TCP has no advantage
against the competitive flows because of no data packet losses
2011/05/12
CQR2011
17
Conclusion
Characteristics of QoS-Guaranteed TCP on Real Mobile
Terminal in Wireless LAN
 QoS-TCP is likely to guarantee a target bandwidth in mobile
wireless environments
Analysis of performance evaluation results
 QoS-TCP cannot defeat competitive TCP in bit error free
simulations
 QoS-TCP in real situation is effective to guarantee
throughput
2011/05/12
CQR2011
18
Thank you for your attention!!
2011/05/12
CQR2011
19