Академический Документы
Профессиональный Документы
Культура Документы
Topic 6: TDMA FDMA CDMA Objective: Explain and Analyze Multi-access Communication Protocols . (Refer T1-Ch.4; T2-Ch. 4 & 6 + Class Notes)
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access
08/23/11
08/23/11
Scheduling
Random access
Polling: take turns Request for slot in transmission schedule Token ring Wireless LANs
08/23/11
Scheduling: Polling
Inbound line
Data to M
Host computer
Outbound line
3 M
Stations
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 5
Scheduling: Token-Passing
Ring networks
token
Data to M token
Random Access
Multitapped Bus Crash!!
08/23/11
Wireless LAN
AdHoc: station-to-station Infrastructure: stations to base station Random access & polling
08/23/11
Scale
How much traffic can be carried? How many users can be supported?
Current Examples:
Design MAC to provide wireless DSL-equivalent access to rural communities Design MAC to provide Wireless-LAN-equivalent access to mobile users (user in car travelling at 130 km/hr)
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access
08/23/11
Delay-Bandwidth Product
Coordination in sharing medium involves using bandwidth (explicitly or implicitly) Difficulty of coordination commensurate with delaybandwidth product
08/23/11
10
Case 1
Case 2
A
A detects collision at t = 2 tprop
B B
08/23/11
11
Station B needs to be quiet tprop before and after time when Station A transmits; so, Waiting time=2tprop R transmission bit rate (Note is different from Effe. Throughput, Reff .) L bits/frame; Sending station requires X=L/R secs for transmitting its frame. Effe. Throughput is Actual rate (bps) at which info. sent over channel Efficiency is Normalized Maximum throughput (= Reff./R). Normalized delay-b/w product (a) = ratio of 1-way delay b/w product to avg. frame length.
Efficiency = max
Normalized Delay-Bandwidth Product
08/23/11
L 1 1 = = = L + 2t prop R 1 + 2t prop R / L 1 + 2a
L 1 = = R bits/secon d L / R + 2t prop 1 + 2a
a=
t prop L/R
08/23/11
13
1 Efficiency = 1 + 2a
CSMA-CD (Ethernet) protocol:
1 Efficiency = 1 + 6.44a
Token-ring network
If a<<1, then efficiency close to 100% As a approaches 1, the efficiency becomes low
1 Efficiency = 1 + a
a= latency of the ring (bits)/average frame length
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 14
Distance
100000 km 3.33 x 1006 3.33 x 1007 3.33 x 1008 Global area network
Max size Ethernet frame: 1500 bytes = 12000 bits Long and/or fat pipes give large a
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 15
08/23/11
Parameters
08/23/11
17
E[T]/X
1 Load
08/23/11
m a x
At low arrival rate, only frame transmission time At high arrival rates, increasingly longer waits to access channel Max efficiency typically less than 100%
18
Transfer delay
Dependence on Rtprop /L
E[T]/X
a > a a a
Transfer Delay
1 max Load
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 19
mx a
08/23/11
20
traffic onto a shared line Packets are encapsulated in frames and queued in a buffer prior to transmission Central control allows variety of service disciplines
Statistical Multiplexing & Random Access allows sharing of a broadcast Multiplexing concentrates bursty MAC
medium Packets are encapsulated in frames and queued at station prior to transmission Decentralized control wastes bandwidth to allow sharing
A Shared Medium
Buffer
R bps
C Input lines
08/23/11
R bps
C Input lines
21
Output line
System Performance Transfer Delay Packet/frame Loss Efficiency & Throughput Priority, scheduling, & QoS
A Shared Medium
Buffer
R bps
C Input lines
08/23/11
R bps
C Input lines
22
Output line
Arrival Model
Independent frame interarrival times: Average 1/ Exponential distribution Poisson Arrivals
Load = /
Stability Condition: <1
Infinite Buffer
No Blocking
(From Appendix A)
08/23/11
In M/G/1 model, a frame arriving to an empty multiplexer begins transmission immediately In many MACs, there is a delay before transmission can begin M/G/1 Vacation Model: when system empties, server goes away on vacation for random time V
1 2
R/M
M stations do not interact Poisson arrivals /M fr/sec Constant frame length L bits Transmission time at full rate X=L/R Station bit rate is R/M Neglect guardbands Transmission time from station L/(R/M)=M(L/R) =MX M times longer Load at one station: = ( / ) X= = X
08/23/11
...
26
The delay increases in proportion with M, the number of stations Allocated bandwidth to a given station is wasted when other stations have data to send
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 27
08/23/11
TDMA
28
MX E[TTDMA ] = MX + +X 2(1 ) 2
Better than FDMA & CDMA Total Delay still grows proportional to M
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 29
08/23/11
30
ALOHA
Wireless link to provide data transfer between main campus & remote campuses of University of Hawaii Simplest solution: just do it
A station transmits whenever it has data to transmit If more than one frames are transmitted, they interfere with each other (collide) and are lost If ACK not received within timeout, then a station picks random backoff time (to avoid repeated collision) Station retransmits frame after backoff time First transmission t0-X t0 t0+X Retransmission t t0+X+2tprop Time-out
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 31
Backoff period B
t0+X+2tprop + B
Vulnerable period
08/23/11
ALOHA Model
Definitions and assumptions X frame transmission time (assume constant) S: throughput (average # successful frame transmissions per X seconds) G: load (average # transmission attempts per X sec.) Psuccess : probability a frame transmission is successful
S = GPsuccess
X Prior interval
X frame transmission
Any transmission that begins during vulnerable period leads to collision Success if no arrivals during 2X seconds
32
08/23/11
Abramsons Assumption
What is probability of no arrivals in vulnerable period? Abramson assumption: Effect of backoff algorithm is that frame arrivals are equally likely to occur at any time interval G is avg. # arrivals per X seconds Divide X into n intervals of duration =X/n p = probability of arrival in interval, then G = n p since there are n intervals in X seconds
Throughput of ALOHA
S = GPsuccess = Ge
0.2 0.18 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0
0 0. 00 78 12 5 0. 01 56 25 0. 03 12 5 0. 06 25 0. 12 5 0. 25
2 G
e-2 = 0.184
Collisions are means for coordinating access Max throughput is mx = 1/2e (18.4%) at a G=0.5 This corresponds to a total arrival rate of exactly one frame / vulnerable period. Bimodal behavior:
Small G, SG Large G, S0
08/23/11
Collisions can snowball and drop throughput to G zero 2 or more arrivals in vulnerable period result ECE C394 Communication Networks collision. 34 Topic 7: ALOHA, Polling, Random Access
0. 5
Slotted ALOHA
Time is slotted in X seconds slots. Frames are assumed to be constant and to occupy 1 time slot. Stations synchronized to frame time. All stations keep track of Tx time slots & initiate Tx at the beginning of one time slot. Stations transmit frames in first slot after frame arrival Backoff intervals in multiples of slots
B t t0 +X+2tprop + B
Only frames that arrive during prior X seconds collide (No collisions in in t0+X from to). Vulnerable period is now X sec. long (not 2X)
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 35
Vulnerabl eperiod
08/23/11
36
...
Reservation mini-slots X-second slot
...
Reservation protocol allows a large number of stations with infrequent traffic to reserve slots to transmit their frames in future cycles Each cycle has mini-slots allocated for making reservations Stations use slotted Aloha during mini-slots to request slots
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 37
Reduce BW wastage by avoiding Tx. that are certain to collide. A station senses the channel before it starts transmission
If busy, either wait or schedule backoff (different options) If idle, start transmission Vulnerable period is reduced to tprop (due to channel capture effect) When collisions occur they involve entire frame transmission times If tprop >X (or if a>1), no gain compared to ALOHA or slotted ALOHA
Station A begins transmission at t=0
A
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 38
08/23/11
Sensing
39
0.53
0.45 0.16
Better than Aloha & slotted Aloha for small a = 0.01 Normalized prop. Delay: a = a =0.1 tprop /X
64
a=1
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access
08/23/11
40
0.81
0.51
0.14
2 4 8 32 16 64
a=1
Higher maximum throughput than a = 0.1 1-persistent for small a G Worse than Aloha for a > 1
- CSMA schemes improve over ALOHA schemes by reducing vulnerable period from one- or two-frame transmission times to a single prop. delay tprop . - In both ALOHA and CSMA schemes, collisions involve entire frame Transmission.
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access
08/23/11
41
Monitor for collisions & abort transmission Stations with frames to send, first do carrier sensing After beginning transmissions, stations continue listening to the medium to detect collisions If collisions detected, all stations involved stop transmission, reschedule random backoff times, and try again at scheduled times In CSMA collisions result in wastage of X seconds spent transmitting an entire frame CSMA-CD reduces wastage to time to detect collision and abort transmission
08/23/11
42
B B
A
A detects collision at A t= 2 tprop -
It takes 2 tprop for, station A, to find out if channel has been captured. Station that detects collision transmits a short jamming signal to ensure that others know of collision before aborting Tx & the back-off algo is used to schedule a future resensing time
08/23/11
43
Channel can be in 3 states: busy, idle, contention period where multiple stations attempt capturing channel. Assumptions Collisions can be detected and resolved in 2tprop
Time slotted in 2tprop slots during contention periods Assume n busy stations, and each may transmit with probability p in each contention time slot Once the contention period is over (a station successfully occupies the channel), it takes X seconds for a frame to be transmitted It takes tprop before the next contention period starts.
CSMA-CD Model
(a)
Busy
Contention
Busy
Idle
Contention
Busy
08/23/11
Time
44
Contention Resolution
How long does it take to resolve contention? Contention is resolved (success) if exactly 1 station (how to identify this 1 station?) transmits (& so n-1 stations do not is when n=1) in a slot:
To find max. achievable throughput, assume that stations use the value of p that maximizes P success . By taking derivative of Psuccess we find max occurs at p=1/n 1 1 1 1
Psuccess = np(1 p ) n 1
max success
= n (1 ) n n
n 1
= (1 ) n
n 1
CSMA/CD Throughput
Busy Contention Busy Contention Busy Contention Busy
At maximum throughput, systems alternates between contention periods and frame Time transmission times
max
where:
R bits/sec, L bits/frame, X=L/R seconds/frame a = tprop /X If a is much smaller (say 0.01), higher meters/sec. speed of light in medium (ie.,94%) throughput is d meters is diameter of system achievable. 2e+1 = 6.44
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 46
mx a
0.6
Slotted ALOHA
0.4
ALOHA
0.2 0 0.01 0.1 1
For small a: CSMA-CD has best throughput For larger a: Aloha & slotted Aloha better throughput
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access
08/23/11
48
08/23/11
49
08/23/11
50
08/23/11
51
Reservations Systems
Centralized systems: A central controller accepts requests from stations and issues grants to transmit
Frequency Division Duplex (FDD): Separate frequency bands for uplink & downlink Time-Division Duplex (TDD): Uplink & downlink time-share the same channel
Central Controller
08/23/11
52
Reservation Systems
Reservation interval Frame transmissions
d Cycle n
d Cycle (n + 1)
Time
1 2
Transmissions organized into cycles Cycle: reservation interval + frame transmissions Reservation interval has a minislot for each station to request reservations for frame transmissions
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 53
Centralized systems: A central controller listens to reservation information, decides order of transmission, issues grants Distributed systems: Each station determines its slot for transmission from the reservation information
08/23/11
54
Example
Initially stations 3 & 5 have reservations to transmit frames
(a) r 3 5 r 3 5 r 3 5 8 r 3 5 8 r 3 t
Station 8 becomes active and makes reservation Cycle now also includes frame transmissions from station 8
(b) 8 r 3 5 r 3 5 r 3 5 8 r 3 5 8 r 3
t
55
08/23/11
max
MkX 1 = = Mv + MkX 1 + v k
08/23/11
56
max =
08/23/11
1 X = X(1+ev) 1 + 2.71v
57
Example: GPRS
General Packet Radio Service
Packet data service in GSM cellular radio GPRS devices, e.g. cellphones or laptops, send packet data over radio and then to Internet Slotted Aloha MAC used for reservations Single & multi-slot reservations supported
08/23/11
58
08/23/11
59
Polling Systems
Centralized polling systems: A central controller transmits polling messages to stations according to a certain order Distributed polling systems: A permit for frame transmission is passed from station to station according to a certain order A signaling procedure exists for setting up order
Central Controller
08/23/11
60
Priority mechanisms
More bandwidth & lower delay for stations that appear multiple times in the polling list Issue polls for stations with message of priority k or higher
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 61
Time to prepare & send polling message Time for station to respond
Walk time: from when a station completes transmission to when next station begins transmission Cycle time is between consecutive polls of a station Overhead/cycle = total walk time/cycle time
Polling messages
1 2 3 4 5
t
2
t
3
t t
4 5
Tc
Assume walk times all equal to t Exhaustive Service: stations empty their buffers Cycle time = Mt + time to empty M station buffers /M be frame arrival rate at a station NC average number of frames transmitted from a station Time to empty one station buffer:
Tstation
Tc = N c X = ( Tc ) X = M M
= X
Tc = Mt + MTstation
08/23/11
Mt = Mt + Tc Tc = 1
63
Exhaustive Service
Cycle time increases as traffic increases, so delays become very large Walk time per cycle becomes negligible compared to cycle time:
Limited Service
MX Mt Efficiency = = Tc
Many applications cannot tolerate extremely long delays Time or transmissions per station are limited This limits the cycle time and hence delay Efficiency of 100% is not possible
MX 1 Efficiency = = MX + Mt 1 + t / X
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access
08/23/11
Transmit mode
Delay
Delay From device To device Ready station inserts its frames Reinserts free token when done
Ready station looks for free token Flips bit to change free token to busy
08/23/11
65
Single-token operation
Free token inserted after last bit of the busy token is received back Transmission time at least ring latency If frame is longer than ring latency, equivalent to multi-token operation
Single-Frame operation
Free token inserted after transmitting station has received last bit of its frame Equivalent to attaching trailer equal to ring latency
08/23/11
66
Definition
: ring latency (time required for bit to circulate ring) X: maximum frame transmission time allowed per station
Multi-token operation
Assume network is fully loaded, and all M stations transmit for X seconds upon the reception of a free token This is a polling system with limited service time:
max
MX 1 1 = = = + MX 1 + / MX 1 + a / M
max =
MX
+ M max{(X, }
1 max{1, a} + a/M
Single-token operation
Effective frame transmission time is X+ ,therefore
max =
08/23/11
68
1.2 1
Maximum throughput
Single frame 0 operation 0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4 4.4 4.8
a <<1, any token reinsertion strategy acceptable a 1, single token reinsertion strategy acceptable a >1, multitoken reinsertion strategy necessary
08/23/11 ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 69
Application Examples
Single-frame reinsertion
IEEE 802.5 Token Ring LAN @ 4 Mbps
Multitoken reinsertion
IEEE 802.5 and IBM Ring LANs @ 16 Mbps FDDI Ring @ 50 Mbps
CSMA-CD
Quick transfer and high efficiency for low delay-bandwidth product Can accommodate large number of bursty users Variable and unpredictable delay
08/23/11
71
Polling
Generalization of time-division multiplexing Provides fairness through regular access opportunities Can provide bounds on access delay Performance deteriorates with large delay-bandwidth product
ECE C394 Communication Networks Topic 7: ALOHA, Polling, Random Access 72
08/23/11