CS 536 Park

IEEE 802.11 MAC

CSMA/CA with exponential backo
almost like CSMA/CD
drop CD
CSMA with explicit ACK frame
added optional feature: CA (collision avoidance)
Two modes for MAC operation:
Distributed coordination function (DCF)
multiple access
Point coordination function (PCF)
polling-based priority
. . . neither PCF nor CA used in practice
CS 536 Park
CSMA: (i) explicit ACK and (ii) exponential backo
Sender:
MAC (rmware in NIC) receives frame from upper
layer (i.e., device driver)
Goto Backoff procedure
Transmit frame
Wait for ACK
If timeout, goto Backoff procedure
Receiver:
Check if received frame is ok
Wait for SIFS
Transmit ACK
CS 536 Park
Backoff:
If due to timeout, double contention window (CW)
Else wait until channel is idle plus an additional DIFS
Choose random waiting time between [1, CW]
CW is in units of slot time
Decrement CW when channel is idle
Return when CW = 0
CS 536 Park
Timeline without collision:
DATA BO BO DIFS DIFS
SIFS
2way handshake
. . .
. . .
Time
Time
Sender
ACK
Receiver
SIFS (short interframe space): 10 s
Slot Time: 20 s
DIFS (distributed interframe space): 50 s
DIFS = SIFS + 2 slot time
BO: variable back-o (within one CW)
CWmin: 31; CWmax: 1023
CS 536 Park
Time snapshot with Mira-come-lately:
Sue sends to Arnold
Arnold
DATA
ACK
Time
Time
SIFS
Time
Sue
Mira
Mira wants to send a frame
DIFS BO
DIFS
Slot Time
. . .
BO
WAIT
CS = BUSY
. . .
Contention Window CW
DATA
CS 536 Park
Time snapshot with collision (Sue & Mira):
Mira
Sue
Time
SIFS
Time
Time
Arnold
Sue wants to send a frame
Mira wants to send a frame
DATA
SIFS
BACKOFF
BACKOFF
Slot Time
WINNER
NO
ACK
ACK
NO
BO
BO DIFS
DIFS
DATA
. . .
. . .
DATA
COLLISION
CS 536 Park
MAC throughput and collision (ns simulation):
0
1
2
3
4
5
6
3.5 4 4.5 5 5.5 6 6.5
T
h
o
u
g
h
p
u
t

(
M
b
/
s
)
Offered Load (Mb/s)
node 2
node 5
node 10
node 20
node 30
node 50
node 100
0
10
20
30
40
50
60
70
3.5 4 4.5 5 5.5 6 6.5
C
o
l
l
i
s
i
o
n

P
r
o
b
a
b
i
l
i
t
y

(
%
)
Offered Load (Mb/s)
node 2
node 5
node 10
node 20
node 30
node 50
node 100
CS 536 Park
MAC throughput:
experiment: iPAQ running Linux
0
1
2
3
4
5
6
4 4.5 5 5.5 6 6.5 7 7.5 8
T
h
r
o
u
g
h
p
u
t

(
M
b
/
s
)
Offered Load (Mb/s)
Node 2
Node 5
Node 10
Node 12
Node 16
CS 536 Park
Additional issues with CSMA in wireless media:
Hidden station problem:
Mobile Mobile Mobile
A B C
Hidden Station Problem
(1) (2)
(3)
(1) A transmits to B
(2) C does not sense A; transmits to B
(3) interference occurs at B: i.e., collision
CS 536 Park
Exposed station problem:
Mobile Mobile Mobile
A
Mobile
B C D
(1) (1) (2)
Exposed Station Problem
(1) B transmits to A
(2) C wants to transmits to D but senses B
C refrains from transmitting to D
omni-directional antenna
CS 536 Park
Solution: CA (congestion avoidance)
RTS/CTS reservation handshake
Before data transmit, perform RTS/CTS handshake
RTS: request to send
CTS: clear to send
SIFS SIFS SIFS
CTS
RTS
Sender
Receiver
reservation handshake
20B
14B 14B
29B 2347B
same as before
. . .
. . .
Time
Time
DIFS DIFS BO
ACK
DATA
CS 536 Park
Hidden station problem: RTS/CTS handshake clears
hidden area
Mobile Mobile Mobile
A B C
RTS
CTS
CTS
RTS/CTS Handshake
"clears the area"
RTS/CTS perform only if data > RTS threshold
why not for small data?
. . . feature available but not actively used
CS 536 Park
Additional optimization: virtual carrier sense
transmit connection duration information
stations maintain NAV (network allocation vector)
decremented by clock
if NAV > 0, then do not access even if physical CS
says channel is idle
CS 536 Park
PCF (point coordination function):
support for real-time trac
Periodically inject contention free period (CFP)
after BEACON
Under the control of point coordinator: AP
polling
PIFS (priority IFS):
SIFS < Slot Time < PIFS < DIFS
CS 536 Park
Properties of PCF:
BEACON period is not precise
has priority (PIFS < DIFS) but cannot preempt
DCF
During CFP services stations on polling list
delivery of frames
polling: reception of frames
must maintain polling list: group membership
Uses NAV to maintain CFP
BEACON: separate control frame used to coordinate
BSS
time stamp, SSID, etc.
CS 536 Park
IEEE 802.11 wireless LAN standard:
ratied in 1997: 1/2 Mbps using either DSSS or FHSS
11 bit chip sequence
uses IEEE 802 address format along with LLC
4 address elds for forwarding/management
uses 2.42.4835 GHz ISM band in radio spectrum
ISM (industrial, scientic and medical): unlicensed
IEEE 802.11b ratied: 5.5/11 Mbps using DSSS only
less coding overhead: good for low BER
BER (bit error rate) and FER (frame error rate)
others: e.g., IEEE 802.11a and 802.11g at 54 Mbps
5.7255.85 vs. 2.42.4835 GHz band
both use OFDM
Bluetooth, 802.16, etc.; uncertain future . . .
CS 536 Park
WLAN: ad hoc vs. infrastructure mode
a.k.a. why ad hoc, in general, is a bad idea
why. . .
Access Point
F1
F1
F1
F1
F1
F1
Mobile
Mobile
Mobile
Mobile
Mobile
Mobile
Mobile
WLAN: Infrastructure Network
F1
F1
F1
F1
F1
F1
Mobile
Mobile
Mobile
Mobile
Mobile
Mobile
Mobile
Mobile
F1
F1
F1
F1
F1
WLAN: Ad Hoc Network
Two key reasons:
nothing to do with wireless
i.e., common to wireline networks
double duty
specically to do with wireless LANs
CS 536 Park
Consider corridor conguration: linear chain
wireline
wireless
... ...
... ...
n/2+1 n/2
e
connectivity-wise: equivalent
Assume:
there are n nodes
link bandwidth: B Mbps
every node picks a random destination to talk
data rate of each node: 1 Mbps
consider middle link e = (n/2, n/2 + 1)
how much trac must go through e?
inherent load or stress on link e
CS 536 Park
Fixing direction (left or right, i.e., half-duplex):
average load on e: L(e) = n/4
note: still linear in n (i.e., L(e) = O(n))
To satisfy trac demand:
bandwidth on e: B = n/4 Mbps
else: cannot send or must buer
Main observation: under ad hoc mode in both wireline
& wireless networks, individual link bandwidth must in-
crease with system size n
due to forwarding duty!
not scalable w.r.t. system size
CS 536 Park
Remarks:
link e = (n/2, n/2 + 1) is typical
majority of links are near the middle
how does link bandwidth requirement X increase in
2-D grid/lattice conguration?
n
n
cut
when side is n long: total n
2
nodes
hence with

n side: total n nodes

note: link e was a cut dividing into 2 halves
average load on each link in 2-D cut?
CS 536 Park
Average link load: L(e) =

n/4
under assumption of perfect load balancing
note:

n number of links in the cut
Thus in grid topologies, bandwidth requirement increases
as: O(

n)
still not scalable
in wireline networks: use switches/routers
else must upgrade link bandwidth constantly
ad hoc WLANs: additional impact of collision/interference
throughput goes down even more
Gupta and Kumar 99