Multiplexing

CSC 343643

WAKE FOREST
U N I V E R S I T Y Department of Computer Science

Fall 2010

Multiplexing
Often two machines can not utilize the full capacity of a link Need to share the channel among multiple users (multiplexing)
89:; ?>=< A J JJ JJ JJ J$ @ABC GFED / M B u: uu u uu uu 89:; ?>=< C @ABC GFED D t: ttt tt tt 89:; / M / ?>=< E JJ JJ JJ JJ $ 89:; ?>=< F

shared link

Three types of multiplexing Frequency-division multiplexing Time-division multiplexing Statistical time-division multiplexing
E. W. Fulp CSC 343643 Fall 2010 1

Frequency Division Multiplexing

Given a medium with bandwidth larger than a single signal Divide bandwidth to allow multiple carrier signals Carrier frequencies may not overlap (analog signaling) Signal raised in frequency to particular carrier
Channel 1 1

Attenuation factor

Channel 2 1

Channel 2 Channel 1 Channel 3

60 Channel 3 1

64

68

72

Frequency (kHz) Multiplex channel

300

3100

60

64

68

72

Frequency (Hz) original bandwidths

Frequency (kHz) Bandwidths raised in frequency

Multiple signals can be sent simultaneously

E. W. Fulp CSC 343643 Fall 2010 2

channel 1

channel 2

channel 3 f3

f2

f1

Telephone system Consider backbone links carrying multiple connections Twelve 4 kHz channels multiplexed together Multiplexed into the 60 to 108 kHz band, called a group What are other examples of FDM?

E. W. Fulp

fre

qu

en c

tim

channel 4 f4

CSC 343643

Fall 2010

Digital Subscriber Line

Using the existing telephony network for high-speed data and voice Local loop twisted pair has a 1.1 MHz bandwidth potential However only a fraction is actually used Voice requires 300 Hz to 3400 Hz Remove (lter) anything below 300 Hz and above 3400 Hz Asymmetric Digital Subscriber Line (ASDL) has larger bandwidth Asymmetric since higher bit rates downstream than upstream Line is not ltered, 1.1 MHz is available

E. W. Fulp

CSC 343643

Fall 2010

Divides 1.1 MHz bandwidth into 256 channels (three bands) 1. 0 to 4 kHz telephone service (POTS) (1 channel) 2. 25 kHz to 200 kHz upstream (5 channels) 3. 250 kHz to 1000 kHz for downstream (remaining channels)
256 4-kHz Channels Power 0 Voice

25 Upstream Downstream

1100 kHz

Each channel modulation is similar to V.34 QAM used, 15 bits/baud, 4000 baud Actual rate also depends on line length and quality What is the bit rate if 224 channels are available?

E. W. Fulp

CSC 343643

Fall 2010

Local exchange/end office

ADSL
POTS transmission + switching network

POTS
LTU
0 4 kHz

NT Low-pass filter POTS splitter Low-pass filter

0 4 25

1100

kHz

High-pass filter

High-pass filter STB of TV or PC

ADSL modem

25

1100

kHz

ADSL modem

ADSL-Lite
POTS LTU

Video servers

Low-pass filter

0 4 25

500 kHz

NT

High-pass filter
ADSL termination network

STB or PC ADSL modem High-pass ADSL-lite filter + modem

The internet

POTS = plain old telephone service NT = network termination ADSL = asynchronous digital subscriber line

E. W. Fulp

CSC 343643

Fall 2010

Cable TV Networks
Uses FDM to transmit multiple channels concurrently Each channel is allocated 6 Mhz (8 MHz in Europe)
Coaxial cable bandwidth Downstream/forward channels

Analog TV channels (6/8 MHz per channel )

Unused

Unused

5 MHz

88/110

300/450/550

900 MHz

Note: Normally, the band of frequencies from 88 110MHz are used for radio broadcasts.

Upstream is 5 - 42 MHz and downstream is above 550 Hz Use QAM-64 per channel What is the maximum data rate per channel?

E. W. Fulp

CSC 343643

Fall 2010

Time Division Multiplexing

FDM requires analog circuits (not amenable to computers) Time Division Multiplexing (TDM) is used by digital electronics However, local loops are analog TDM works by interleaving multiple signals over time

channel 4 channel 3 channel 2 channel 1 channel 4 channel 3 channel 2 channel 1

E. W. Fulp

fre

qu

en

cy

Generic TDM Description

A n signals mi (t), i = 1...n are to be multiplexed Input data may be briey buered Buers scanned to produce a composite digital stream
buffer m1(t)

tim

CSC 343643

Fall 2010

m2(t)

1 2 3

...n

...

server

mn(t)

Data rate of the composite stream

n

rc
i=1
E. W. Fulp

ri
CSC 343643 Fall 2010 9

Digital Carrier Systems

TDM is used in telephone networks, voice is 4 kHz 8000 PCM samples are taken per second at 8 bits/sample therefore a voice channel is 64 kbps What is the time per sample? 24 voice channels are multiplexed together (T1)
193 Bit frame (125 sec)

Channel 1

Channel 2

Channel 3

Channel 4

Channel 24

1 0

Bit 1 is a framing code

7 Data bits per channel per sample

Bit 8 is for signaling

E. W. Fulp

CSC 343643

Fall 2010

10

T1 frame consists of 24 8 = 192 bits plus 1 extra bit for framing, total of 193 bits every 125 sec What is the bit rate of a T1? If a T1 is used for data (non-voice), then only 23 channels are used, 24th is for synchronization T1 can be multiplexed into higher carriers
4 T1 streams in 40 51 4:1 62 73 1.544 Mbps T1 6.312 Mbps T2 44.736 Mbps T3 274.176 Mbps T4 6 5 4 32 10 7:1 6:1 1 T2 streams out 7 T2 streams in 6 T3 streams in

E. W. Fulp

CSC 343643

Fall 2010

11

TDM Link Control

The T1 description does not have headers, trailers, CRCs, etc... Control mechanisms (data-link) are not needed Flow control Multiplexed data rate is xed Multiplexer and demultiplexer operate at same rate Therefore ow control is not needed What if a user transmits more or less than subscribed? Error control Why retransmit an entire T1 frame? Flow and error control are handled on a per channel basis
E. W. Fulp CSC 343643 Fall 2010 12

Multiplexing Analogy
3-lane highway used by employees of 3 companies (IBM, Cisco, Nortel) FDM Assign a lane to each company Employees must use lane assigned to their company Can use assigned lane at any time TDM Assign specic times for highway use 1. IBM employees 5:00am - 6:00am 2. Cisco employees 6:01am - 7:00am 3. Nortel employees 7:01am - 8:00am Employees use entire highway, but only during assigned time Any problems with FDM or TDM?

E. W. Fulp

CSC 343643

Fall 2010

13

 STDM Any employee can use any lane at any time Works only if there is no peak time, based on averages If all employees use highway at 8:00am ... Assume every day trash builds-up along the highway, and you (acting a governor) have the right to ne the company (instead of the individual employee). Is it easy to identify the oending employee group using the dierent multiplexing techniques?

E. W. Fulp

CSC 343643

Fall 2010

14

Statistical Time Division Multiplexing

Computer trac (data) is often bursty, not a constant bit stream On/o oriented (consider telephone conversations)
1

0.5 amplitude

-0.5

-1 0

5000 time

10000

15000

Using TDM, if a station is idle transmit what? Statistical TDM exploits the fact that digital data is often bursty Dynamically allocate time-slots (channels) on demand What is the disadvantage?

E. W. Fulp

CSC 343643

Fall 2010

15

Generic STDM Description

Slots are not assigned to a particular source
t3 t3 t2
a4

t2

t1

t0
a1

t1

t0

b4

b3 a4 b2 a3 b1 a2

a3 a2 a1

TDM t3 server t2 t1 t0

b4 b3

b2

b1

b4 b3 a4 b2 b1 a3 a2 a1

STDM

Assume arriving frames join a queue Frames will be buered until transmitted Addresses are required more overhead Why are addresses required?

E. W. Fulp

CSC 343643

Fall 2010

16

STDM Performance
Sum of the average input rates is less than the multiplexed line However, the sum of the input peaks is greater If all inputs transmit at the peak... Typically buering is done on the input side This temporarily stores the data until it is served Smoothes the data stream by removing burstiness Buering does add possible transmission delays Analysis is based on queueing theory Interested in the average packet delay Delay is based on the arrivals and the server

E. W. Fulp

CSC 343643

Fall 2010

17

A Simple Model
buffer arrivals server depatures

Packets arrive at the system according to arrival pattern Packets can be from one or multiple sources (multiplexer) If the buer is nite and full, the packet is dropped Packets are serviced in a FIFO order We are interested in a relationship between the following N (t) the number of packets in the system at time t (t) the number of arrivals in interval 0 to t Ti the time spent in the system by packet i

E. W. Fulp

CSC 343643

Fall 2010

18

Littles Theorem
An intuitive notion of the typical number of packets in the system 1 Nt = t
t 0

N ( )d

the time average of N ( ) up to time t Nt will change over time, but it will approach a steady state N = lim Nt
t
1 0.9 average number of packets 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0 500 1000 time 1500 2000 2500

E. W. Fulp

CSC 343643

Fall 2010

19

 Similarly, we can make the same claim about the arrivals t = (t) t

the average arrival rate up to time t The steady state arrival rate is = lim t
t
0.5 0.45 average arrival rate 0.4 0.35 0.3 0.25 0.2 0.15 0.1 0 500 1000 time 1500 2000 2500

E. W. Fulp

CSC 343643

Fall 2010

20

The time average of the packet delay up to time t is Tt =

(t) i=1 Ti

(t)

the average time spent in the system per packet up to time t The steady state time average packet delay is T = lim Tt
t

These quantities (N , , and T ) are related by a simple formula, also known as Littles Theorem N = T It expresses the natural idea that crowded systems (large N ) are associated with long delays (large T ) Valid for dierent types of queueing systems...
E. W. Fulp CSC 343643 Fall 2010 21

Graphical Proof
Littles Theorem is really an accounting identity
number of arrivals number of departures 4 3 2 1 arrivals

N(t)

T2 packet 2 T1 packet 1

departures

time

Suppose the system is initially empty N (0) = 0 Let (t) be the number of arrivals and (t) be the number of departures at time t The number in the system at time t is N (t) = (t) (t)
E. W. Fulp CSC 343643 Fall 2010 22

The area between the graphs of (t) and (t) is

t 0

N ( )

(1)

Let t be any time at which the system is empty N (t) = 0, then the same area is equal to
(t)

Ti
i=1

(2)

Setting equation (1) equal to equation (2) and dividing by t 1 t This equals Nt = t Tt
t 0

1 N ( ) = t

(t)

Ti =
i=1

(t) t

(t) i=1

Ti

(t)

E. W. Fulp

CSC 343643

Fall 2010

23

Applications
The signicance of Littles Theorem is its generality Holds for almost any queueing system that reaches steady state However, it may require information we dont know Given the statistics about the arrivals and departures, we can derive the wait time as well as the queue length Given a certain arrival and departure process, we want the average delay...

E. W. Fulp

CSC 343643

Fall 2010

24

M/M/1 Queue
buffer arrivals server depatures

Consider a single buer and server, assume Interarrival times are exponential with rate > 0 Processing times are exponential with rate > 0 Buer space is innite Average delay is T = 1

Can determine the delay given the average arrival and processing

E. W. Fulp

CSC 343643

Fall 2010

25

M/M/1 Performance
As the utilization approaches 1, performance decreases (unstable) System is stable when < 1 (underload)
M/M/1 Queue 25 average number in the system 10 M/M/1 Queue

20 average delay 0.2 0.4 0.6 utilization 0.8 1

15

10

0 0

0 0

0.2

0.4 0.6 utilization

0.8

Why does queueing increase dramatically when = ? Why does queueing exist when < (underload) condition?

E. W. Fulp

CSC 343643

Fall 2010

26

Average Rate
The arrival pattern of packets (customers) is important We need to also describe the variation
Rate = 2.5 2 2 Rate = 2.5 1.5 1.5

arrival

arrival 1 2 time 3 4

0.5

0.5

0 0

0 0

10 time

15

20

12 10

45 40 35

number of arrivals

8 6 4 2

number of arrivals 1 2 3 interarrival time 4 5

30 25 20 15 10 5

0 0

0 0

5 interarrival time

10

15

So what? Request CSC 790 High-Speed Networks and nd out more...

E. W. Fulp

CSC 343643

Fall 2010

27

Back to TDM and STDM Performance

Results for the M/M/1 queue can be applied multiplexing systems TDM consists of dividing the capacity into n channels Each channel has rate Arrival rate Transmission rate Using Littles Theorem, delay per packet is TT DM = 1
r n

Each stream can be modeled as an M/M/1 queue with

E. W. Fulp

CSC 343643

Fall 2010

28

STDM buers packets from n streams Assume multiplexed in one FIFO buer Total arrival rate is n System is an M/M/1 queue with transmission rate n Average delay is TST DM = TT DM 1 = nn n

1 therefore the STDM system delay is a fraction ( n ) of TDM

E. W. Fulp

CSC 343643

Fall 2010

29

 Example, channel with 128kbps data rate 10 users that transmit packets with independent exponentially distributed lengths of 1000 bits. Each packet is transmitted according to a Poisson process with average rate 10 packets per second. TDM, each user gets a 12.8kbps channel TT DM = STDM all users share the link TST DM = TT DM = 36msec 10 1 = 360msec 12.8 10

E. W. Fulp

CSC 343643

Fall 2010

30

TDM and STDM Comparison ( = 1) 100 90 80 70 average delay 60 50 40 30 20 10 0 0 0.2 0.4 0.6 arrival rate () 0.8 1 TDM STDM

E. W. Fulp

CSC 343643

Fall 2010

31

Multiplexing Results
Preceding example indicates TDM (separate channel) multiplexing results in poor delay Performance is even worse if channels are not assigned proportionally to arrival rate Arrival rates tend to change over time... However, models presented here rely on assumptions that may not be applicable to actual trac TDM/FDM is still used in telephony since voice trac tends to be regular rather than Poisson

E. W. Fulp

CSC 343643

Fall 2010

32