SONATA Switchless Optical Network for Advanced Transport Architecture Project AC351

1
SONATA (Switchless Optical Network for Advanced
Transport Architecture) Project AC351
Modelling and Performance Evaluation of a
National Scale Switchless Based Network
Josep Solé-Pareta, Davide Careglio, Salvatore
Spadaro, Jaume Masip, Juanjo Noguera and Gabriel
JunyentUniversitat Politècnica de Catalunya -
UPCAdvanced Broadband Communications Lab. - CCABA
2
Outline

• Objective and Concept of SONATA
• SONATA Network Controller
• Scheduling Algorithm
• Simulation Environment
• Experiments and Results
• Per Packet Request Approach
• Per Packet-Flow Request Approach
• Conclusions

3
Objective of SONATA

• MAIN OBJECTIVE to define and demonstrate a
  single-layer network for end-to-end optical
  packet connections between a large number of
  terminals

• EXPLOITED CONCEPTS
• packet-by-packet wavelength tuning of
  transmitters and receivers with a central
  wavelength router for interconnections between
  groups of terminals
• a central wavelength conversion stage to improve
  interconnection flexibility
• elimination of complex switching nodes from the
  network ("switchless" network)

4
Switchless Network Concept
network

• wavelength agile terminals (Tx and Rx)

buffers
controller

• Hyper-PON infrastructures

x
T
R
x
x
T
x

• a single wavelength router

R
x
R
x
T
x

• wavelength conversion

R
x
T
x
HyperPON
T
x
R
x

• cell-by-cell wavelength switching at Tx and Rx

T
x
R
x
R
x

• multi-point-to-multi-point connections

T
x
R
x
T
x
HyperPON
l conv. array
HyperPON
l conv. array
5
SONATA Network Controller (I)

• Attached to the PWRN (Passive Wavelength Routing
  Node)
• To access to the Network Controller each PON has
  at least one wavelength assigned
• All terminals of the same PON have to be share
  the wavelength assigned to that PON
• This device is responsible for
• The resolution of access conflicts (TDMA
  protocol)
• The allocation of transmission resources to user
  (SONATA Terminals)
• The configuration of the network connectivity
  using wavelength converters arrays

6
SONATA Network Controller (II)

• The time on each wavelength is divided into slots
  which are organized into frames

• Each terminal asks for the reservation of
• a large amount of slots without particular time
  constraints (datagram traffic)
• an integer number of slots in each frame for the
  whole connection duration (constant bit-rate
  service)

• In case of a large network, a simple resource
  allocation algorithm is required for speed reasons

7
SONATA Network Control (III)

• Simple approach based on the decoupling of the
  time dimension from the wavelength dimension

• Resource allocation problem was split into two
  sub-problems
• Scheduling of user request in the time domain
  given a PON-to-PON channel assignment
• Design of the network connectivity by properly
  assigning wavelengths to PONs (logical topology
  design)

8
Scheduling algorithm (I)

• Implementation of the scheduling algorithm is
  based on a set of matrices, which represents the
  network available resources
• Matrix W Each entry points out to a list of
  records describing the wavelength channels that
  provide connectivity between corresponding
  PON-to-PON pair
• Matrix S indicates, for each wavelength channel,
  which slots of the frame are already assigned
• Matrix Utx indicates for each transmitter (TX)
  which slots of the frame are already assigned
• Matrix Urx indicates for each receiver (RX)
  which slots of the frame are already assigned

9
Simulation Environment (I)

• Simulation HPC platform provided by CEPBA

• Silicon Graphics Origin 2000
• 64 processors MIPS-R10000
• Main memory 8 GBytes
• Memory cache 4 MBytes
• Theoretical peak performance 32 Gflop/s

10
Simulation Environment (II)

• Simulation language TeD (GeorgiaTech)

• Telecommunications Description (TeD) Language
• is designed for execution efficiency and ease of
  use
• supports modularity and model reuse

• TeD language specifications consist of two parts
• MetaTeD a meta language specification to
• provide high-level specification of structure and
  behaviour of networking elements
• structure the simulation to facilitate parallel
  execution
• entities communicating via event scheduling
  mechanisms
• no shared state between entities
• External language coupling to MetaTeD a
  programming language (e.g., C, Java) to
• express exact, detailed, executable
  implementations of higher level specifications

11
Simulation Environment (III)

• Simulation model
• Request generators based on Bernoulli random
  sources.
• The signaling bandwidth bottleneck is avoided
  using more than one signaling channel per PON (Nc
  gt 1)
• Scheduling algorithm module implemented based on
  the above described
• Balanced traffic Uniform distribution of the
  destination PONs (all PONs has the same
  probability to be destination of the traffic)
• Unbalanced traffic Gaussian distribution of the
  destination PONs

1
PON 1
. . .
Nc
Scheduling Algorithm
. . .
1
PON NP
. . .
Nc
12
Simulation Environment (IV)

• Simulation parameters Network Configuration
• Number of PONs, NP 100
• Number of SONATA terminals per PON, NST 1000
• Mini-slots per frame, Ks 1000
• Number of wavelengths used for signaling, Nc 3
  or 10
• Slots per Frame, 100 (frame duration, 1 ms)
• Transmission bit rate per channel, 622 Mbps
• Number of dummy ports used 0, 100, 200, 300 or
  400 (0, 1, 2, 3, 4 additional
  wavelengths per PON-to-PON pair)

13
Simulation Environment (V)

• Simulation parameters implications
• Assuming either the FTTCab or the FTTK/B
  solution, and 200 end-users per SONATA terminal,
  this approach can support 20 million end-users
  (NEU 20 106 End-users)

14
Simulation Environment (VI)

• Simulation parameters Traffic characteristics
• Only Internet Traffic (no connection oriented
  traffic)
• Internet traffic requests modelled by the
  Bernoulli generators
• Data traffic (IP datagrams) offered load per PON,
  p 0.6 - 0.75 - 0.9
• Data traffic destination PONs distribution
• Balanced Uniform (Mean 1/100)
• Unbalanced Gaussian (Mean 50, Standard
  Deviation 20)

15
Experiments and Results

• Scheduling algorithm performance evaluation
• Per packet request approach (requests were issued
  by the IP layer, one per IP packet to be
  transmitted).
• Per packet-flow request approach (requests were
  issued from the TCP layer, one per TCP session).
• Network operation time simulated, 10 sec.
• Execution time required, 36 h.

16
Per Packet Request Approach (I)

• Number of slots required per request 70 of the
  requests would ask for 1 slot and the rest (30)
  would ask for 2 slots
• Network Controller serves the requests in a frame
  by frame basis, i.e., a request can only be
  served if the slots (1 or 2) can be allocated in
  the data frame which is being scheduled at that
  moment. If not, request is lost
• Number of signalling channels, NC 10

17
Per Packet Request Approach (II)

• Performance evaluation
• Request Loss Rate
• Resource Occupancy (Throughput)
• p is the requests offered load (to the signaling
  channels)
• ? is the network offered load

18
Per Packet Request Approach (III)

• Sample of Results using two wavelength converters
  plus the direct wavelength per PON (?max 0.43)
• Balanced traffic

19
Per Packet Request Approach (IV)

• Sample of Results using one wavelength converter
  plus the direct wavelength per PON (?max 0.65)
• Balanced Traffic

Requests Loss Rate
1,E00
1,87E-01
Direct
3,77E-01
2,94E-01
Wavelength
Direct Wavelength
5,70E-02
1,E-02
1,16E-02
Total
3,54E-03
RLR
Direct wavelength
1,E-04
Dummy port
Dummy port
1,E-06
0,5
0,6
0,7
0,8
0,9
1,0
p
r
r
0.32
0.49


0.59

0.65


0.32 0.49

0.59 0.65
20
Per Packet Request Approach (V)

• How to stress the network to test the SONATA
  potential capabilities?
• Increase the number of signalling wavelengths
  (NC)
• With this approach the signalling bandwidth
  easily becomes a bottleneck
• Reduction of the amount of the signalling by
  issuing per packet-flow requests instead of per
  individual packet request
• This option is not realistic since signalling
  cannot be implement at TCP/UDP level, anyway it
  completely avoids the bottleneck-signalling
  problem.

21
Per Packet-Flow Request Approach (I)

• Traffic assumption
• 42 of the traffic is http (35) ftp (7), for
  this traffic
• 20 of flows in average require 2 slots
• 21 of flows in average require 13 slots
• 6 of flows in average require 130 slots
• 0.5 of flows in average require 1300 slots
• 45 of the traffic is irc (23) telnet (7)
  smtp (5) games (5) etc. (5), for this
  traffic
• 100 of flows in average require 1 slot
• The remainder 13 is IP over IP traffic
  (tunneling)
• We considered that this traffic has the same
  composition than the ordinary IP traffic (70 1
  slot and 30 2 slots)
• According this assumptions, the distribution for
  the number of required slots per request that we
  adopted in our simulations was
• 1 (46), 2 (23), 13 (23), 130 (8)

22
Per Packet-Flow Request Approach (II)

• The request are served by allocating the
  requested slots in consecutive data frames
• Number of signalling channel, NC 3

23
Per Packet-Flow Request Approach (III)

• Performance evaluation
• Request Loss Rate
• Resource Occupancy (Throughput)
• p is the requests offered load (to the signaling
  channels)
• ? is the network offered load

24
Per Packet-Flow Request Approach (IV)

• Sample of Results considering four wavelength
  converters per PON (?max 0.86)
• Unbalanced Traffic

Requests Loss Rate
Resources occupancy (Throughput)
Direct wavelength
6,33E-01
100
6,88E-01
1,E00
83,99
4,68E-01
86,24
5,65E-01
3,90E-01
3,01E-01
80,22
Direct wavelength
80
2,96E-01
1,60E-01
2,08E-01
1,E-01
62,83
58,38
8,86E-02
5,28E-02
60
51,91
51,10
Total
4,27E-02
RLR
1,E-02
43,48
Resources occupancy ()
48,30
Direct wavelength 4 Dummy ports
34,83
40
42,09
35,73
1,E-03
32,46
21,95
26,23
4,45E-04
20
9,52
Dummy port 4
2,00E-04
6,69
1,E-04
0,04
0
p 0,35
0,45
0,55
0,65
p 0,35
0,45
0,55
0,65
? 0.36 0.45 0.54
? 0.36 0.45 0.54
25
Per Packet-Flow Request Approach (V)

• Per PON wavelength occupancy distribution
  considering four wavelength converters per PON
• Unbalanced Traffic

Wavelength Distribution Occupancy
Wavelength Distribution Occupancy
?
?
( 0,36 - p 0,42 )
( 0,54 - p 0,63 )
Direct Wavelength
Direct Wavelength
100
100
80
80
60
60
Occupancy ( )
Occupancy ( )
40
40
Total
20
20
Total
Dummy port 4
Dummy port 4
0
0
0
20
40
60
80
100
0
20
40
60
80
100
Destination PON
Destination PON
26
Per Packet-Flow Request Approach (IV)

• Sample Results considering two wavelength
  converters per PON
• Balanced Traffic

Resources occupancy (Throughput)
Requests Loss Rate
Direct Wavelength
1,E00
Direct Wavelength
100
89,06
89,53
6,33E-01
5,62E-01
89,22
4,53E-01
88,22
2,74E-01
80
84,75
Dummy port 1
1,E-01
1,42E-01
80,09
Total
70,21
60
Direct wavelength 1 Dummy port
67,00
1,E-02
RLR
Resources occupancy ()
62,51
1,59E-02
53,75
40
7,50E-03
7,61E-04
27,20
1,E-03
20
Dummy port 2
2,06E-04
Direct wavelength 2 Dummy port
1,82
1,E-04
0
? 0,5
0,6
0,7
0,8
0,9
1,0
? 0,5
0,6
0,7
0,8
0,9
1,0
p 0.35 0.42 0.52 0.63
0.7
p 0.35 0.42 0.52 0.63
0.7
27
Conclusions

• SONATA Network is a switchless single layer
  optical network designed to cover national and
  metropolitan area, which exemplifies a possible
  vision of future optical Internet backbones
• Performance evaluation results obtained by
  simulation show that the SONATA approach is
  feasible and can provide good Resource
  Occupancy and Request Loss Rate figures
• Nevertheless, in order to reach these figures a
  dynamic assignment of the dummy ports wavelength
  to the PON (dynamic logic topology design)
  algorithm has to be used to implement the Network
  Controller