Scenario
STA
STA
STA
STAAP
Server Fixed Network
Server
Server
STA
STAAP
Upwnlink TCP flows
Downlink TCP flows
What is the WLAN cellperformance with bidirectional TCP traffic?
Mobile Communications Networks
3
VoIP over WLANs
• Voice over IP– Economical alternative to
traditional telephony.
• Wireless LANs– Broadband access network to
internet.
• VoIP over WLANs.– A market opportunity.
• Complement/Substitute of current cellular networks.
– Technical limitations.
178,000 new Skype users per day!
August 2003
Mobile Communications Networks
4
VoIP over WLANs
• Voice over IP– Economical alternative to
traditional telephony.
• Wireless LANs– Broadband access network to
internet.
• VoIP over WLANs.– A market opportunity.
• Complement/Substitute of current cellular networks.
– Technical limitations.
178,000 new Skype users per day!
August 2003
Mobile Communications Networks
5
VoIP over WLANs
• Limitations for VoIPoWLAN (and other realtime services)– Protocol overheads.– Capacityvarying channels (multirate).– Random Access MAC.
• Bandwidth penalty due to the distributed access.– Unfairnes Downlink/Uplink.
• Difficult coexistence between heterogeneous (rigid and elastic) flows.– Handoff / Roaming between WLANs.
Mobile Communications Networks
6
Example: Rdata = 2 Mbps, Rbasic = 1 Mbps, VoIP codec: G.729 (8 Kbps)– Downlink/Uplink TCP flows ↑ simultaneous VoIP calls ↓
IEEE Solution: EDCA (Enhanced Distributed Channel Access): • Traffic differentiation at MAC layer + CAC optional.
• EDCA is too conservative.• Static Traffic Differentiation without Admission Control.
VoIP over WLANs
VoIP call
VoIP call
ACK packets
TCP packets
ACK packets
TCP packets
TCP packets
ACK packets
Mobile Communications Networks
7
Multimedia Traffic Flows
• General Classification (queue occupation). – Elastic flows (TCPbased):
• use all available bandwidth, fair with other elastic flows.• saturated source/queue with : B ≈ L/X, ρ =1
– Streaming/Rigid flows (UDPbased): • use only the bandwidth required.• finiteload (unsaturated) source/queue with: B ≈ ρ L/X, ρ ≤ 1
Rigid Traffic (UDP)
User Perception
B
B
• Video/audio streaming• Voice over IP
• Web (HTTP)• FTP, email, ...
Multimedia
Elastic Traffic (TCP)User Perception
MinimumBandwidth
Good
Bad
Better
Worst
Mobile Communications Networks
9
The scenario
AP
MN
MN
MN
MN
MN
MN
MN
MN
MN and AP share the channel using a random access MAC protocol.
B=100 Kbps
B=200 Kbps
B=80 Kbps
B=120 Kbps
B=450 Kbps
B=max. av.
B=max. av.
B=max. av.
Be=max. av.
MN1
MN2
Mobile Communications Networks
10
BUSY
BUSY
BO
IEEE 802.11 MAC Model (DCF)
• Each node is modelled by a M/M/1/K queue with packet arrival rate λ k and service time X.– Assumption: Poisson arrivals and service time exponentially distributed.– The model accounts for: queuing delays, packet losses.
t
t
MN1
MN2
X11/λ 1
X2
packet arrival(HOL)
packet arrival(queued) packet departure
DATA ACK BO DATA ACK
packet departure
X1
EQ1:Queueing delay
BO DATA ACKBUSY BO
backoff suspension
BO
BO
COLLISION
COLLISION
Queue empty
1/λ 2
K
Xλ k
Mobile Communications Networks
11
BUSY
BUSY
BO
IEEE 802.11 MAC Model (DCF)
• Each node is modelled by a M/M/1/K queue with packet arrival rate λ k and service time X.– Assumption: Poisson arrivals and service time exponentially distributed.– The model accounts for: queuing delays, packet losses.
t
t
MN1
MN2
X11/λ 1
X2
packet arrival(HOL)
packet arrival(queued) packet departure
DATA ACK BO DATA ACK
packet departure
X1
EQ1:Queueing delay
BO DATA ACKBUSY BO
backoff suspension
BO
BO
COLLISION
COLLISION
Queue empty
1/λ 2
K
Xλ k
Mobile Communications Networks
12
DCF EDCA
BUSY
BUSY
BO t
t
MN1
MN2
X11/λ 1
X2
packet arrival(HOL)
packet arrival(queued) packet departure
DATA ACK BO DATA ACK
packet departure
X1
EQ1:Queueing delay
BO DATA ACKBUSY BO
backoff suspension (succ.)
BO
BO
COLLISION
COLLISION
Queue empty
1/λ 2
tBO BUSY
MN3
backoff suspension (coll.)
empty slot
AIFS
BEB
TXOP
Mobile Communications Networks
13
IEEE 802.11e MAC Model (EDCA) : TXOP• Multiple packet transmission each time a MN/AP gets the channel
– The overall network throughput increases (S=Data/Time [bps])– Less wasted time in random access.
BO DATA ACK
BO DATA ACKbusy
BO busy
BO DATA ACK
BO DATA ACKbusy
BO busy
DATA ACK
BO DATA ACK
GAIN
BO DATA ACKBO busy
BO DATA ACK
Packet arrival
Packet arrival
Packet arrival (Queued)
Packet arrival (Queued)
Packet arrival
Packet arrival (Queued)
Packet arrival
Packet arrival (Queued) Service Time (X)
MN 1
MN 2
MN 1
MN 2
GAIN
Service Time (X)
t
t
t
t
No TXOP
TXOP
Mobile Communications Networks
14
IEEE 802.11e MAC Model (EDCA) : TXOP
• Each user is modeled as an M/G[1,B]/1 queue (bulk service time queue)– TXOP (Burst length): maximum consecutive number of frames transmitted at each
successful attempt.
0 1 2 i=B j K
• Stationary (equil.) distribution.• Blocking Probability.• Queue occuppation.• Queuing delay.
• Departure distribution.• Average TXOP length• Average TXOP duration
Renewal Analysis
Mobile Communications Networks
15
IEEE 802.11e MAC Model (EDCA): AIFS
• AIFS
BOBUSY BO tBO BUSY
Extra blocked slots due to AIFS
BUSYBUSYBUSY
BOBUSY BO tBO BUSYBUSY BUSY BUSY
Aggregate load from other nodes
MN1
MN1
Service Time ↑Transmission Prob. ↓
Mobile Communications Networks
16
IEEE 802.11e MAC Model (EDCA): EB, p
• Expected BackOFF slots
• Cond. Collision Probability
• Prioritizing a flow: TXOP ↑ AIFS ↓ CWmin↓
• This model is able to capture the impact of the other flows with different MAC parameters.
Mobile Communications Networks
17
Wireless Scenario
• SingleHop AdHoc network.• No hiddenterminals and ideal channelconditions.• Channel access based on the DCF (BA and RTS/CTS).• Two types of flows: streaming (UPDlike) and elastic (TCPlike).
3
2
n1
n
i1
Mobile Communications Networks
18
Results
• Singlecell WLAN (adhoc)– Two types of flows:
• S1 (streaming ~ unsaturated)– Bs = 100 Kbps– L = 400 bytes
• E1 (elastic ~ saturated)– Be (network dependant)– L = 1500 bytes
• Default WLAN MAC parameters– 0 S1 flows.
• Solution (1):– AIFS ↑ (elastic flows, A=3)
• 8 S1 flows– The TCP throughput is reduced.
S1 saturation point
Mobile Communications Networks
19
Model Validation
• Singlecell WLAN (adhoc) – Two types of flows:
• S1 (streaming ~ unsaturated)– Bs = 100 Kbps– L = 400 bytes
• E1 (elastic ~ saturated)– Be (network dependant)– L = 1500 bytes
• Solution (2):– TXOP S1 ↑ (=4)
• 7 S1 flows.– CWmin,E1 ↑ (=64)
• 6 S1 flows.
Mobile Communications Networks
21
Problem statement: Providing QoS
• Flowlevel Admission Control. • Adaptive selection of MAC parameters.
Mobile Communications Networks
22
Problem statement: Providing QoS
• Flowlevel Admission Control. • Adaptive selection of MAC parameters.
random access MAC(Performance)=f(MAC parameters, number of nodes, traffic profile each node)
non-linear relation
maximize performance (under the QoS)
Mobile Communications Networks
23
The available bandwidth estimation problem
nn
BandwidthBandwidth
Bandwidthpenalty
Flow rejected Flow rejected
... how to decide if a flow can be admited? We have to compute or to testif the new flow is possible.
Ideal Channel Sharing Random Access Channel Sharing
... the bandwidth penalty due to the distributed access depends on the number of flows active and the traffic profile of each flow.
Availablebandwidth Available
bandwidth
Mobile Communications Networks
27
Problem Statement
Distribution of VoIP flows in a cell (G.711 codec)
While total cell capacity is around 12 calls when all of them transmit at 11Mbps…
…this capacity falls with any change from fast to slow flows
Mobile Communications Networks
29
VoIP Capacity in a mesh network
MP MP MP MP
VoIP calls (G.279)
Single channel
Channel Contention (MAC)
hidden nodes
D. Niculescu. S. Ganguly, K. Kim, R. Izmailov; “Performance of VoIP in a 802.11 Wireless Mesh Network”. IEEE Infocom 2006, Barcelona, Spain.
2 Mbps
Mobile Communications Networks
30
VoIP Capacity in a mesh network
MP MP MP MP
VoIP calls
Multiple channels(Two channels)
Channel Contention (MAC) ↓
hidden nodes ↓
MP MP MP MP
VoIP calls
NO Channel Contention (MAC)
NO hidden nodes
A
BMultiple channels(Three channels)
Mobile Communications Networks