Impairment Aware RWA Algorithms Based on Prediction Concepts - Workshop Vilanova July 7

Page created by Bob Richardson
 
CONTINUE READING
Impairment Aware RWA Algorithms Based on Prediction Concepts - Workshop Vilanova July 7
UNIVERSITAT POLITÈCNICA DE CATALUNYA

           Impairment Aware RWA
        Algorithms Based on Prediction
                   Concepts
                                        Eva Marín-Tordera
                          Advanced Network Architectures Lab (CRAAX)
                            Universitat Politècnica de Catalunya, UPC
                                    Vilanova i la Geltrú, Spain

                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                                         Index

                                   •    Introduction
                                   •    Part 1
                                         – Prediction-Based Routing (PBR)
                                         – Maximum Transmission Distance (MTD)
                                         – MINCOD routing algorithm
                                         – PR-MTD RWA algorithm
                                         – PR-MTD performance evaluation
                                   •    Part 2
                                         – Q Personik’s factor
                                         – Planning Phase – Deterministic Algorithm
                                         – Analysis Phase – PR-Q RWA algorithm
                                         – Simulation framework
                                         – PR-Q performance evaluation
                                   •    Conclusions

                                                                                      2
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                              Introduction

                                     •      Our research group have been actively working in the
                                            design of RWA algorithms that utilize inaccurate network
                                            state information.
                                     •      This experience has been transferred to the area of IA-
                                            RWA.
                                           1. Taking into account the Physical Impairments (PI) as a
                                              constraint more.
                                           2. Taking into account that the Physical information can be
                                              inaccurate
                                     •      Work done in collaboration with: Telecom Italia,
                                            Corecom, Politecnito di Milano and Pirelli Labs

                                                                                                         3
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                              Part 1

                                   • Proposal of IA-RWA algorithm PR-MTD taking
                                     into account:
                                       – Maximum Transmission (MTD) Distance as physical
                                         impairment.
                                       – Inaccurate Network state information (wavelength
                                         availability).
                                       – Removing update messages with wavelength
                                         availability changes.
                                   • Work developed in Nobel 2 Project
                                   • Publications: ICTON 2007 and ECOC 2007
                                                                                        4
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                      Prediction Based Routing

                                     Prediction Based Routing (PBR)
                                                                                              • Prediction → Read 2-bit counter.
                                   • Motivation: The network state information • Update two-bit counter
                                     (wavelength availability, regenerator usage and PI → By increasing if connection is
                                     values) used by usual routing algorithms might not
                                     be completely accurate.                                  blocked.
                                         – Assuming source routing, source nodes need → By decreasing if connection is
                                         update messages with information about wavelength
                                         availability, changes in PLI values and regeneration established.
                                         capabilities in nodes.                                          PTsource-destination
                                   • Goal:    Proposing a routing mechanism based            Lightpath                      2 bit counter Prediction
                                     on prediction issues named Prediction-Based                 i        2 bit counter
                                                                                                                                  00 (0d) Not blocked
                                     Routing (PBR) aiming at:                                                                     01 (1d) Not blocked
                                         – Reducing the routing Inaccuracy problem effects                                        10 (2d) Blocked
                                                                                                                                  11 (3d) Blocked
                                         – Reducing (even removing) the amount of update
                                         messages

                                   • Main features: The PBR does not extract the
                                     required network state information from the
                                     update messages, instead it infers this information
                                     from the behavior of previous connection
                                     requests.

                                                                                                                                               5
                                    Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT       Maximum Transmission Distance (MTD)
                                   •       Full transparency is not however always achievable in long
                                           distance networks.
                                   •       In a semi-transparent network routes are divided into transparent
                                           sub-routes between signal regenerators.
                                   •       MTD (Maximum Transmission Distance) is the maximum distance
                                           a signal can be routed on without any regeneration1.
                                   •       Wavelengths are classified into three classes (gold, silver and
                                           bronze), each one characterized by a different value of the MTD2 .
                                                                                   4500

                                                                                   4000

                                                                                   3500
                                                                             [km
                                                                  disttance[km]

                                                                                   3000

                                                                                   2500
                                                                 distance

                                                                                                                        Gold class (d > 3500 km)
                                                                                   2000
                                                                                                                        Silver class (3000
COMPUTER ARCHITECTURE DEPARTMENT                  MINCOD Routing Algorithm
                                   • Minimum Coincidence and Distance                    100                                   250

                                     Algorithm (MINCOD) routing algorithm                      50      100                      150
                                                                                                                                                  100

                                     precomputes k routes:
                                          – Computes the end to end paths
                                          considering the routes that have less                150    200          100         50            50     50
                                          shared links and minimum distance.                                       50
                                                                                                                                    50                   50

                                   •   MINCOD steps to compute k-routes:                 100          200    100
                                                                                                                                            200
                                          – Selectsthe shortest path (in distance)                                       100
                                          – Computes the Minimum Shared Link                         125
                                          (MSL) of the rest of paths
                                                                                      Shortest Paths
                                                                                                                                          MINCOD Paths
                                               MSL =Distance* (SL +1)                    (Kms)
                                                                                                                                            (Kms, SL)
                                       Distance: distance of the path in Kms.           1-6-8-12-13 (400)
                                       SL: number of links shared between each path                                                          1-6-8-12-13
                                       and the previous selected paths.                  1-7-12-13 (450)                                     (400,SL=0)
                                                                                        1-2-8-12-13 (600)                                      1-7-12-13
                                   •   Select the path with minimum MSL               1-2-3-4-9-11-14-12-13                                   (450, SL=1)
                                   •   Repeat the last two steps (k-2) times              (725)                                          1-2-3-4-9-11-14-12-13
                                                                                      1-2-3-5-9-11-14-12-13                                   (725, SL=2)
                                                                                          (750)
                                                                                                                                                              7
                                   Workshop Vilanova July 7
Prediction Routing according MTD
                                                         PR-MTD
COMPUTER ARCHITECTURE DEPARTMENT

                                                                      K number of routes
                                                                      W number of wavelengths
                                   •   The PR-MTD utilizes k          SRi number of subroutes of route i
                                       previously computed
                                                                      For i=0 to K-1
                                       routes by means of the                For j=0 to W-1
                                       MINCOD routing                              NotReachable=0;
                                       algorithm                        For t=0 to SRi-1
                                                                           If Distance(t) ≥ MTD(j) NotReachable=1;
                                                                        EndFor
                                   •   It selects the first             If NotReachable==0
                                                                        If wavelength j available in all the links of route i
                                       lightpath with two-bit              ROUTE=i;
                                       counter lower than 2,               LAMBDA=j;
                                       output link availability and           break; // the lightpath is assigned
                                                                        EndIf
                                       accomplishing the MTD            EndIf
                                       constraint in each one of            EndFor
                                       the sub-routes.                EndFor

                                                                                                                          8
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT          PR-MTD Performance Evaluation
                                   Network Scenario:
                                   • 2 fibres per link and 40 wavelengths per fibre.
                                   • Wavelengths are divided into three classes, gold, silver and bronze, with MTD of 4000, 3500 and
                                     3000 km respectively.
                                   • Nodes Madrid, Barcelona, Paris, Dublin, Milan, Frankfurt, Amsterdam, Prague, Stockholm and
                                     Athens act as source and destination; and nodes, Frankfurt, Amsterdam, Vienna, Milan, Prague and
                                     Warsaw have regenerators.
                                   • Comparing PR-MTD with SP-LL and MINCOD-LL (SP-LL and MINCOD-LL with periodically updating)

                                                                                                                         SP-LL(1)
                                                                                                               0,4
                                                                                                                         SP-LL(5)
                                                                                                                         SP-LL(10)
                                                                                                              0,35       2_MINCOD-LL(1)
                                                                                                                         2_MINCOD-LL(5)

                                                                                   % of Blocked Connections
                                                                                                               0,3       2_MINCOD-LL(10)
                                                                                                                         2_PR-MTD
                                                                                                              0,25

                                                                                                               0,2

                                                                                                              0,15

                                                                                                               0,1

                                                                                                              0,05

                                                                                                                0
                                                                                                                     0    0,1        0,2     0,3     0,4   0,5       0,6
                                                                                                                                           Erlangs
                                                                                                                                                                 9
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                              Part 2

                                   • Proposal of IA-RWA algorithm PR-Q taking into
                                     account:
                                         – Q Personik’s factor as physical impairment
                                         – Inaccuracy in physical network state information (Q
                                           value).
                                   • Work developed in Nobel 2 Project
                                   • Performance evaluation enhanced to be
                                     submitted to IEEE Networks Special Issue

                                                                                           10
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                                     Q Personik’s factor

                                   • Work developed by M. Quagliotti and W.Erangoli from Telecom Italy and Corecom.
                                   • Q Personik’s factor is a Quality factor that aggregates different physical impairment effects

                                                         Qend = a0 + a1OSNR end + a2 NL tot
                                                        OSNR is the optical signal to noise ratio expressed in dB

                                                     NL takes into account the effects of non-linearity

                                                                 Q factor of this lightpath is 22.6 dB
                                             Link 1: 128 km                                                               Link 3: 580 km
                                                                     Link 2: 298km

                                    Genova              Milano                           Pisa                                                      Rome
                                        a     1     2
                                                          b      1    2     3        4
                                                                                             c       1       2       3           4
                                                                                                                                                   e
                                                                                                                                     5     6   7

                                      node                                                       Amplifier distance: max 85 km
                                              Line amplifier                                     Span distance
                                                                             pre-amplifier
                                         booster

                                                                                                                                                          11
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT      Planning Phase- Deterministic Algorithm

                                   • Network planning, made according to a static traffic
                                     pattern with a static (deterministc) RWA and
                                     Regenerator Placement algorithm.
                                   • Network is equipped with one regenerator per node and
                                     one system (40 wavelengths) per link.
                                        – Each permanent connection requests is routed following the
                                          Shortest Path and First Fit RWA.
                                        – Regeneration is needed if Qlightpath < Qthreshold (17 dB)
                                        – After serving each request, the nodes with no more free
                                          regenerators are provided with an extra regenerator and the
                                          links with no more free wavelengths are provided with an
                                          additional system.
                                   • Additional     check          that      removes        unused
                                     regenerators/systems.
                                                                                                  12
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT        Planning Phase- PanEuropean Results

                                   • 2 traffic matrixes are used:
                                        – a uniform matrix →fullmesh, 378 bidirectional connections
                                        – polarized matrix →Traffic from year 2007, 741 bidirectional
                                          connections (Nobel 2 Data)
                                   • The line system installed on the links of the network are always 40
                                     channels DWDM system at 10 Gbit/s
                                   • Results from this static simulations give us for each case:
                                        – Number of systems installed in each link
                                        – Number of regenerators in each node

                                                                                         Required full
                                                                         Required      line system (40 λ
                                                         Traffic type
                                                                        Regenerators         each)

                                                         Uniform            129               56

                                                        Year 2007           197               79

                                                                                                           13
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT     Analysis Phase- Predictive Algorithm
                                   • Prediction Route according to the Q factor (PR-Q):
                                        – Based on the PBR mechanism.
                                        – Takes into account the inaccuracy on the Q value.
                                   • PR_Q selects the first route and corresponding
                                     wavelengths (wavelength conversion is assumed in
                                     nodes with regeneration) fulfilling:
                                        – Qsub-route > Qthreshold
                                        – 2 bit-counters of each sub-route (and wavelengths) < 2
                                        – There is wavelength availability in all the sub-routes of the
                                          lightpath
                                   • If the selected ligthpath is:
                                        – Blocked → 2-bit counter increased
                                        – Established → 2 bit counter decreased

                                                                                                     14
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                         Simulation Framework
                                   • Q_route is the Q factor of a lightpath computed by the IA-RWA using a
                                     set of stored parameters, independently on the fact that the values are
                                     right or wrong
                                   • Q_actual is the real Q factor in the network
                                   • Q_design is the Q factor considered by the planning algorithm
                                   3 Simulation cases for Dynamic Traffic:
                                   • Perfect Knowledge and Perfect matching (PKPM)
                                            Q_route= Q_actual= Q_design
                                   • Perfect Knowledge and Imperfect Matching (PKIM)
                                            Q_route= Q_actual ≠Q_design
                                    Q_route and Q_actual have pessimistic values (Q_design – 2 dB)
                                   • Imperfect Knowledge and Imperfect Matching (IKIM)
                                            Q_route ≠ Q_actual
                                    Q_actual= Q_route – 2dB Q_actual (real) is worse than Q used by IARWA

                                                                                                               15
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                                    Performance Evaluation
                                                              20%
                                                                    PR-Q
                                                                    Deterministic
                                           Blocking Probabi   15%

                                                                                                  Fullmesh Traffic
                                                              10%
                                                                                                  Dynamic Traffic:
                                                              5%
                                                                                                  1 Erlang between
                                                                                                  every source
                                                              0%                                  destination →
                                                                         PKPM       PKIM   IKIM
                                                                                                  378 Erlangs
                                                              10%
                                                                    PR-Q
                                                                    Deterministic
                                                              8%
                                                                                                  Year 2007 Traffic
                                           Blocking Probabi

                                                              6%
                                                                                                  Dynamic Traffic:
                                                              4%
                                                                                                  517 Erlangs
                                                              2%

                                                              0%
                                                                         PKPM       PKIM   IKIM

                                                                                                                 16
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT                                  Performance Evaluation
                                                                             Fullmesh Traffic

                                                             18

                                                             16

                                                             14
                                   Bloking probability (%)

                                                             12       PKPM PR-Q
                                                                      PKIM PR-Q
                                                             10       IKIM PR_Q
                                                                      PKPM Deterministic
                                                             8
                                                                      PKIM Deterministic
                                                             6        IKIM Deterministic

                                                             4

                                                             2

                                                             0
                                                                  1                  2                    3              4
                                                                           Offered traffic source-destination (Erlang)
                                                                                                                             17
                                   Workshop Vilanova July 7
COMPUTER ARCHITECTURE DEPARTMENT               Conclusions and Future Work
                                   • We have introduced the Prediction Based Routing Mechanism in
                                     the Impairment Aware RWA algorithms.
                                          –      Taking into account the inaccuracy in the wavelength availability
                                          –      Taking into account the inaccuracy in the Physical Impairment
                                                 information

                                   • Routing algorithms inferred from the PBR performs better than
                                     usual IA-RWA algorithms in case of inaccurate information
                                     conditions.
                                   • In DICONET project we continue working developing IA-RWA
                                     algorithms:
                                          –      Not only based on PBR
                                          –      For the planning and the analysis cases
                                          –      Specifically designed for each PI (or different PIs)

                                                                                                                 18
                                   Workshop Vilanova July 7
You can also read