Broadcast and Select

Broadcast&Select WDM networks - 1

Optical Networks: from fiber transmissionto photonic switching

Broadcast-and-Select Networks

Fabio Neri and Marco MelliaTLC Networks Group Electronics Department

e-mail: [email protected]://www.tlc-networks.polito.it/

[email protected] tel. 011 564 [email protected] tel. 011 564 4173


Broadcast-and-select networks

No routing, but full connectivity: information from a source is broadcasted to all receivers; receivers select the information directed to them, and discard the rest

Similar to traditional LAN operation

All nodes have visibility on all network traffic, but do not need to process (switch) all of it

The aim is to limit the bandwidth processed by network nodes, finding the proper balance between photonic and electronic technology


Broadcast-and-select networks Typical topologies: star, bus and ring

It is possible to have star of stars, ring of rings, etc.

starcoupler

1

7 3

8 2

56 4


Passive couplers Passive component, to couple or split signals in different

fibers

Can be fiber based, or realized in waveguides Can be wavelength selective When

= , the input power on one fiber is (as a first

approximation) halved on each output fiber (3 dB coupler) When 1 (0.9 - 0.95), we have a tap

output 1

output 2input 2

input 1

O1 =

I1 + (1-) I2

O2 = (1-) I1 + I2



12345678

12345678

3-dB coupler

Using a passive star coupler it is possible to build the broadcast star topology

overall n/2 log2 n 22 (3 dB) couplers log2 n 22 devices crossed on all paths: same power loss

for every node pair


A bus topology requires 2n 22 couplers Losses along the path are typically larger

(linear with n) Losses are different for each node pair

1 2 3 4 5 6 7 8

3-dB coupler




Each node is usually attached to two fibers: one to transmit, one to receive

W WDM channels are available Tx and Rx operate on a single WDM channel at

a time (to reduce electronic bandwidth) It is possible to observe collisions and contention

Collision: two or more transmitters transmit on the same channel at the same time

Contention: a single receiver must tune to two or more channels at the same time

We need a Medium Access Control (MAC) protocol


Broadcast-and-select networks Nodes can be equipped with one or more tx and rx

devices, which may be tunable or fixed Tunable txs and rxs are more expensive (and tunable rxs

usually cost more than tunable txs) Connectivity may be limited due to components or

complexity constraints

For example, if node i has a fixed tx on i and a fixed rx on |i-1|N , a ring logical topology results

Traffic will be then routed using multi-hop paths, going through a number of intermediate nodes, where OEO conversion is performed


Broadcast-and-select networks With limited connectivity, a logical topology is built

over the (broadcast star) physical topology For example: 2 fixed tx/rx per node allow to build a

shuffle topology

1

2

3

4

5

6

7

8

1

2

3

4

1234

5

786

91011

12

13

1516

14

1234

5

786


Broadcast-and-select networks Different resource allocation strategies can be adopted

when the traffic pattern is relatively stable (flow duration much larger than propagation delays), or when a dynamic, packet by packet, network control is necessary

Often time is slotted, and statistical time multiplexing is adopted

Tuning time of tx/rx may be a non-negligible and must be taken into proper account

One (or more) channel can be devoted to signaling (almost necessary in broadcast stars )

Slot synchronization does not come for free, since the slot time is small (guard times must be small compared to the slot duration)

Bit synchronization must be faced as always


Slot synchronization


Synchronization problems Varying propagation delays must be equalized (ranging) Slot phase (when the slot starts) and frequency (how long does

it last) information must be distributed; chromatic dispersion and tuning latencies must be taken in proper account using guard times at slot boundaries

The bit frequency (not the bit phase) can be broadcasted (to simplify the receiver design), or acquired at each receiver

The bit phase must be acquired at receivers Burst mode receivers are necessary: each rx can receive from

a different source in different time slots (we no longer have point-to-point channels); therefore bit synchronization and decision threshold must be acquired for each new reception


Slotted Aloha / Slotted Aloha Broadcast-and-select star network; N nodes;

W


Slotted Aloha / Slotted Aloha When node x has to transmit a packet:

It selects according to some criteria (e.g. at random) a transmission channel T

It transmits a control frame (containing destination node and chosen data channel) using c

The data frame is transmitted on T in the next minislot Every node

Keeps listening to the control channel When a transmission to its address is detected, it tunes the rx on

channel T There can be collisions (on c and on T ) and contention (on T ) This is called tell-and-go approach (data is assumed to be

received after large propagation delays)


Slotted Aloha / Slotted Aloha


Slotted Aloha / Slotted Aloha Two tx/rx pairs per node are necessary:

One fixed pair locked to the control channel One tunable pair for data channels

To avoid useless transmissions on the data channels after collisions on control and data channels, it is possible to use a wait-and-see approach: do not transmit data until the control frame is received back (after a propagation delay); we get:

higher throughput higher access delays

To reduce hardware costs, it is possible to use only one tunable tx/rx pair (with TDM implementation of the control channel)

Variable-size packets can be accommodated by specifying the packet length (in minislots) in control packets


DT-WDMA Broadcast-and-select star network; N nodes;

W=N channels plus a control channel on c Time is divided into minislots (signaling) and

slots (data), with N minislots per data slot

tdata N=6

Nodes are equipped with a fixed transmitter and a tunable receiver for data, plus a fixed tx/rx pair on c

t

control

data t


DT-WDMA Due to transmitter-dedicated channels, no

collision is possible on data channels No collision on the control channel, due to TDMA

access There could be contention on the data channel;

we assume that all nodes use the same algorithm to solve these contentions, so that no explicit ack is necessary

The destination id is the only information carried by the control frame (source and data channel are known due to the TDMA control access)


Scheduling protocols It is possible to increase performance keeping the same DT-

WDMA scheme, but delaying transmission of data frames If data frame allocation is decided by all nodes with the

same rules after having received all reservations, contentions can be avoided

data

control

end-to-end delay

reservationsfor slot X

slot X

Access to the control channel can become random (instead of deterministic TDM) to improve scalability


Larger delays at low loads Larger maximum throughput

Performance of scheduling


Scheduling protocols The previous schemes are distributed: no central control

node, no delay in collecting reservations They can be viewed as extensions of traditional and

successful LAN MAC protocols to a multi-channel setup If access delays are increased, and network resources

are allocated to signaling, it is possible to collect all requests in a central node (or in all network nodes), and compute a schedule for transmissions in the next slot, or next set of slots (frame)

In case of a centralized scheduler, the outcome of the scheduling algorithm must be returned to network nodes

The scheduling (or resource allocation) algorithm can operate on the knowledge of the traffic request matrix

R[i,j]


Switches and B&S networks Broadcast-and-select networks are

equivalent to packet switches Nodes in the network correspond to input/output

ports (linecards) of the switch The central star corresponds to the switching

fabric The same equivalence holds between MAC

algorithms and scheduling algorithms in switches similarities for input buffering, output buffering,

and speed-up can be easily recognized


Output buffer switch

switching fabricoutput 1

output P

input 1a1

input PaN

..

..

..

..

..

..

xP

x1


Output buffering In the packet switch: the switching fabric

and the output card buffers must operate at the aggregate speed Limited scalability

In the B&S network: output buffering is equivalent to have W=N channels, and N receivers in each node

In both systems, it is possible to show that this complexity is not needed


Input buffer switch

input 1a1

input PaN

..

..

..

output 1

output P

..

..

..

scheduler

xP

x1

D

switching fabric


Input buffering Considering W=N WDM channels, the B&S

network behaves as a rearrangeably non- blocking crossbar (if W


Input buffer switch with VOQ

aP

input P

a1

input 1

scheduler

output 1

output P

..

..

..

..

..

..

x1,1x1,P

a1,1

a1,P

1

P

xP,1xP,P

aP,1

aP,P

1

P

D

switching architecture


Matching probleminput output input output


Scheduling problem Matching algorithms on bipartite graphs

are used

requests maximummatching

maximalmatching

Maximum matching has a complexity O(N2.5) Requests can be weighted (priority, queue

length, age, ) leading to weighted matching


Scheduling By increasing the access delay, it is possible to solve a

time/frequency scheduling problem considering several slots in a time frame

Input is the traffic request matrix (which can be static in case of persistent requests, or dynamic)

Constraints: no more than one transmission per TX and no more than one transmission per RX in each time slot; tuning latencies may have to be taken into account

Utility: find the minimum number of slots (minimum frame size) to satisfy all requests, or minimize losses for a fixed frame

Off-line or on-line traffic scenarios; single hop or multihop strategies


Scheduling example (I) A traffic rate matrix (node to node rates are normalized

to channel capacity) is given (N=4 nodes):0.0 0.7 0.1 0.10.1 0.0 0.5 0.2 0.5 0.1 0.0 0.40.4 0.2 0.4 0.0

Each row sum is the total tx rate of one node: must be less than or equal to the number of transmitters available at that node (we assume one)

Each column sum is the total rx rate of one node: must be less than or equal to the number of receivers available at that node (we assume one)


Scheduling example (II) A frame size F must be decided; longer frames give

finer granularity but more complexity; we take F=10 Rates must be translated in slots per frame; in our

case:0 7 1 11 0 5 2 5 1 0 44 2 4 0

A time-wavelength plan must be built, subject to constraints (no more than one packet can be received and transmitted by a node in a slot)

We assume that i delivers information to receiver i


Scheduling example (III) Largest demands are served first (scanning the traffic matrix

by rows) The 4 slots from 4 to 3 cannot be accommodated (the red

slots cannot be used due to constraints); we move forward in time four transmissions from 2 to 3

3 3 3 3 3 4 4 4 41 1 1 1 1 1 12 2 2 2 2

3 3 3 31 2 3 4 5 6 7 8 9 10 1 2 3 4 5

1234


Scheduling example (IV) The 2 slots from 4 to 2 cannot be accommodated

(the red slots cannot be used due to constraints); we move forward in time two transmissions from 1 to 2

3 3 3 3 3 4 4 4 41 1 1 1 1 1 14 4 4 4 2 2 2 2 22 2 3 3 3 31 2 3 4 5 6 7 8 9 10 1 2 3 4 5

1234


Scheduling example (V) The slots from 3 to 2 cannot be accommodated; we

swap in time the last two transmissions to 1, 2 and 3

3 3 3 3 3 4 4 4 4 21 1 1 1 4 1 1 1 44 4 4 4 2 2 2 2 2 12 2 1 3 3 3 31 2 3 4 5 6 7 8 9 10 1 2 3 4 5

1234


Scheduling example (VI) The final scheduling (just one of the several

possible) is a sequence of switching configurations or of input/output permutations

3 3 3 3 3 4 4 4 2 4 3 3 3 3 1 1 1 1 4 1 1 1 4 3 1 1 1 1 4 4 4 4 2 2 2 2 1 2 4 4 4 4 2 2 1 3 3 3 3 2 2 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5

1234


Broadcast-and-select testbeds Several B&S testbeds were described since the early 90s:

Lambdanet [Bellcore 1990]: 18

1.5 Gb/s with 2 nm channel spacing Testbed NTT [NTT 1993]: 100

622 Mb/s with 10 GHz channels spacing

Rainbow I [IBM 1990]: 32

300 Mb/s with 1 nm channel spacing Rainbow II [IBM 1996]: 32

1 Gb/s with 1 nm channel spacing SONATA [E.C. ACTS 1999]: 800

622 Mb/s with 6.25 GHz (0.05 nm) channel spacing

Today the interest in B&S networks for metro networks has decreased

Passive Optical Networks (PONs) take a similar approach for high-speed access and traffic concentration


Considering ring topologies With an approach similar to B&S star networks,

consider a ring topology for applications in metropolitan area networks

Rings help to distribute synchronization information, permit spatial reuse of resources, ease the design of distributed access schemes, ease fault recovery, but introduces larger losses

Usually N users, WN WDM channels, tunable transmitters and fixed receivers

Examples: CORD, Daisy+SR3, Hornet, Wonder Recently interconnected WDM rings have been

proposed (e.g. IST DAVID project)


Arrayed Waveguide Grating (AWG) Generalization of Mach-Zehnder interferometers Wavelength-routing capabilities

Largely used as WDM mux/demux

AWG

11, 21, 31, 4112, 22, 32, 4213, 23, 33, 4314, 24, 34, 44

11, 22, 33, 4414, 21, 32, 4313, 24, 31, 4212, 23, 34, 41


AWG-based networks AWGs have very interesting features to

implement single-hop interconnections using tunable TXs and (tunable) RXs

TT1TT2

TTN

RX1RX2

RXN

AWG No broadcasting Max total

bandwidth = N RXs can be fixed


AWG-based network with couplers

Max total bandwidth = N2 Packet scheduling needed Tunable RXs (and TXs) needed

AWG+

+

TTTTTTTT+

RXRXRXRX+

TTTTTTTT

TTTTTTTT

RXRXRXRX+

RXRXRXRX+

couplers couplers


AWG-based networks with couplers

The space equivalent is a 3-stage setup, with a full mesh in the central stage

Boxes are non-blocking switches (e.g. crossbars)

TXTXTXTX

RXRXRXRX

TTTxTXTX

TTTTTTTX

RXRXRXRX

RXRXRXRX


AWG-based networks

One channel available between any pair of groups of terminals

AWG+

+

TTTTTTTT+

RXRXRXRX+

TTTTTTTT

TTTTTTTT

RXRXRXRX+

RXRXRXRX+


AWG-based networks

AWG+

+

RXRXRXRX+

TTTTTTTT

TTTTTTTT

RXRXRXRX+

-converter

WDM mux WDM demux

One extra channel available between any pair of groups of terminalsScheduling is more complex

Diapositiva numero 1Broadcast-and-select networksBroadcast-and-select networksPassive couplersBroadcast-and-select networksBroadcast-and-select networksBroadcast-and-select networksBroadcast-and-select networksBroadcast-and-select networksBroadcast-and-select networksDiapositiva numero 13Synchronization problemsDiapositiva numero 15Diapositiva numero 16Diapositiva numero 17Slotted Aloha / Slotted AlohaDiapositiva numero 19Diapositiva numero 20Scheduling protocolsPerformance of schedulingDiapositiva numero 23Switches and B&S networksOutput buffer switchOutput bufferingDiapositiva numero 27Input bufferingInput buffer switch with VOQMatching problemDiapositiva numero 31Diapositiva numero 32Scheduling example (I)Scheduling example (II)Scheduling example (III)Scheduling example (IV)Scheduling example (V)Scheduling example (VI)Diapositiva numero 39Considering ring topologiesArrayed Waveguide Grating (AWG)AWG-based networksAWG-based network with couplersAWG-based networks with couplersAWG-based networksAWG-based networks

Broadcast and Select

Documents

networks nodes

optical networks

networks typical topologies

single wdm channel

w wdm channels

node i

passive star coupler

n losses