Optical Packet Switching

Optical Packet Switching DEISNet Gruppo Reti di Telecomunicazioni http://deisnet.deis.unibo.it WDM Optical Network Legacy Networks Edge Systems WDM Links λ 1 λ 2 λ 3 λ 4 Core Nodes 2 1

Wavelength Routing Network Legacy Networks Edge Systems WDM Links Optical Cross-Connects 3 Optical Packet-Switched Network Legacy Networks Edge Systems WDM Links Optical Packet Routers 4 2

Optical Packet Switching Long-term alternative to the optical circuit switching techniques currently under development (e.g. Wavelength Routing, MPλS) Availability of the optical resource at packet level efficient use of the bandwidth Optical Packet Format: Payload Header Guard Band No O/E/O payload conversion needed at the core nodes, only header conversion may be performed 5 Optical Packet Router Architecture Wavelength Demux and Header Extraction 1 N Wavelength Conversion n Non-Blocking, All-Optical Space Switch + + Optical Buffer Switching fabric n Wavelength Mux and Header Insertion 1 N Switch Control Logic 6 3

Contention resolution in OPS networks Typical problem of packet-level switching Resolution techniques available in OPS networks: wavelength domain: contending packets transmitted on separate wavelengths of the same WDM link λ 1 λ 2 λ 3 λ 4 t 0 7 Contention resolution in OPS networks Typical problem of packet-level switching Resolution techniques available in OPS networks: wavelength domain: contending packets transmitted on separate wavelengths of the same WDM link time domain: contending packets delayed by optical buffers λ 1 t 0 t 0 +D 8 4

Contention resolution in OPS networks Typical problem of packet-level switching Resolution techniques available in OPS networks: wavelength domain: contending packets transmitted on separate wavelengths of the same WDM link time domain: contending packets delayed by optical buffers space domain: contending packets forwarded to different links, according to a given adaptive, multi-path routing strategy F 1 to node i F 2 to node j t 0 9 Contention resolution in OPS networks Typical problem of packet-level switching Resolution techniques available in OPS networks: wavelength domain: contending packets transmitted on separate wavelengths of the same WDM link time domain: contending packets delayed by optical buffers space domain: contending packets forwarded to different links, according to a given adaptive, multi-path routing strategy Performance of OPS contention resolution schemes traditionally evaluated in terms of packet loss rate packet latency Effects on the packet stream are also important packet sequence delay jitter 10 5

Optical Buffer 0 D t 0 (B -1)D t 0 t 0 +D t 0 +2D t 0 +(B -1)D Realized with B Fiber Delay Lines (FDL): the delay must be chosen at packet arrival packets are delayed until the output wavelength is available available delays are consecutive multiples of the delay unit D (different choices are also possible) packets are lost when the buffer is full, i.e. the required delay is larger than the maximum delay achievable D M = (B -1)D 11 The FDL-gap problem Using asynchronous, variable-length optical packets is efficient to carry one or more IP datagrams or packets from heterogeneous legacy networks Assumption: FIFO queuing, i.e. packets cannot overtake one another Due to the packets variable length and the finite delay granularity, when packets are queued a time gap between the end of a packet and the beginning of the following one is present (0 < gap size < D) t 0 D GAP t 0 +3D 12 6

The Excess Load During the gap: the server is idle the queued packet must wait for the delay to expire the result is some waste of the available output bandwidth The gap can be seen as an artificial increase of the length of the previous packet it results in an excess load at the output Real Length Excess length D 13 Choosing the Buffer Delay Unit D is directly related to time resolution of the FDL buffer maximum delay achievable (buffer size) Performance Degradation Short buffer Large gaps D 14 7

Single wavelength case Simulation VS. approximate Markov model assuming a finite queue Packet loss probability Too many FDLs Wavelength MUX required D (normalized to the average packet length) 15 Wavelength and Delay Selection (WDS) At packet arrival Given the output fiber (from routing table lookup) Forwarding algorithm must determine the output wavelength the required delay Example: 4 delays 4 wavelengths λ 1 λ 2 λ 3 λ 4 t 0 t 0 +D t 0 +2D t 0 +3D t 0 +4D WDS problem: how to assign the best wavelength to the packet? 16 8

Queuing Model of a WDM Optical Buffer For each output fiber: n identical servers represented by the n wavelengths n logical FDL buffer in parallel, one for each wavelength (realized with a single set of FDLs operated in WDM) Packet λ 1 λ 2 λ 3 λ 4 17 WDS policies Gap-filling techniques may be too complex for the optical packet switching time-scale λ 1 λ 2 λ 3 λ 4 t 0 t 0 +D t 0 +2D t 0 +3D t 0 +4D 18 9

WDS policies Gap-filling techniques may be too complex for the optical packet switching time-scale Simpler policies based on wavelength availability, with increasing intelligence Pursuing an analytical approach seems very complex simulation study λ 1 λ 2 λ 3 λ 4 t 0 t 0 +D t 0 +2D t 0 +3D t 0 +4D 19 WDS policies D-type: delay oriented, aiming at minimizing the queuing time choice of the first available wavelength G-type: gap oriented, aiming at minimizing the gaps between queued packets choice of the wavelength with the closest queued packet G-type λ 1 λ 2 λ 3 λ 4 D-type t 0 t 0 +D t 0 +2D t 0 +3D t 0 +4D 20 10

Single-node WDS performance 1 1.0e-1 1.0e-2 1.0e-3 1.0e-4 1.0e-5 Packet Loss Probability Simple Round-Robin D-type G-type Input Load per wavelength = 0.8 Uniform random traffic 4 I/O fibers 16 wavelengths/fiber 8 FDLs 1.0e-6 0 0.5 1 1.5 2 2.5 3 3.5 4 D (normalized to the average packet length) 21 Role of time and wavelength domains D-type vs. G-type G-type D (normalized to the average packet length) Number of wavelengths per fiber 22 11

Synchronous network, slotted operation Asynchronous, variable-length legacy packets are split and inserted into a number of slots ( slot size = T ) Alternatives for the optical packet format: Fixed-Length Packet (FLP): each slot is routed independently Slotted Variable-Length Packet (SVLP): each train of slots is routed altogether Incoming traffic net load = ρ real load = ρ F >ρ FLP SVLP T real load = ρ S <ρ F 23 SVLP performance Slot size normalized to the average packet length D (normalized to the average packet length) Number of wavelengths in the buffer 24 12

QoS differentiation in the OPS node Due to FDL buffering constraints, traditional priority queuing and scheduling techniques are not feasible QoS differentiation at the OPS node level possible through resource partitioning any K wavelengths are reserved to HP traffic based on the actual occupancy e.g. K=2 LP packets are allowed as long as more than 2 wavelengths are available LP HP λ 1 λ 2 λ 3 λ 4 t 0 t 0 +D t 0 +2D t 0 +3D t 0 +4D λ 1 λ 2 λ 3 λ 4 t 0 t 0 +D t 0 +2D t 0 +3D t 0 +4D 25 Routing strategies in OPS networks Path computation algorithms provide: a default path, i.e. the shortest path (e.g. in terms of number of hops) a number of alternative paths, with equal or higher hop count than the default one The routing strategy may use: Shortest-Path Routing (SPR) only static routing tables not using any alternative path Multi-Path Routing (MPR) dynamic routing tables alternative paths used to relieve congestion on the default one Decisions to be taken by MPR strategy how many alternative paths should be considered how they should be dealt with by WDS policies 26 13

Dynamic MPR strategies (1) Strategies applying WDS on one of the alternative paths only when the default path is congested: a path is considered congested when, on the corresponding output link, all the wavelengths are busy and there are no places left in the optical buffer SAP (Shortest Alternative Paths) beside the default link, an alternative set of routes is considered, including any other shortest path different from the default one n-sap (n-shortest Alternative Paths) besides the default link, n alternative sets of routes are considered, corresponding to paths with up to n 1 hops more than the shortest one 27 Dynamic MPR: 2-SAP Alternative paths available toward the packet destination PKT alternative path 1 DESTINATION default path alternative path 2 28 14

Dynamic MPR: 2-SAP Packets are sent through the default path, as long as this one is not congested PKT DESTINATION WDS 29 Dynamic MPR: 2-SAP In case the default path is congested, the best wavelength is chosen by the WDS policy on one of the alternative paths WDS PKT DESTINATION congestion 30 15

Dynamic MPR: 2-SAP In case the default path is congested, the best wavelength is chosen by the WDS policy on one of the alternative paths congestion DESTINATION congestion PKT WDS 31 Dynamic MPR strategies (2) Strategies applying WDS not on a single link, but on an entire set of links sharing the wavelengths belonging to different paths SSP (Shared Shortest Paths) WDS is performed over all the wavelengths on any shortest path link, including the default one n-ssp (n-shared Shortest Paths) WDS is performed over all the wavelengths on any link corresponding to paths with up to n 1 hops more than the shortest one 32 16

Dynamic MPR: SSP WDS performed on any shortest path link WDS DESTINATION 33 Dynamic MPR: 2-SSP WDS performed on any shortest and 2nd-shortest path link WDS DESTINATION 34 17

Dynamic MPR strategies performance Simulation of European network (15 nodes, 24 links) Dynamic MPR effective when non-shortest paths are used Network-wide Packet Loss Probability 1e-03 1e-04 1e-05 1e-06 SPR SAP 2-SAP SSP 2-SSP 35 QoS differentiation at the routing level Integration of QoS management into dynamic MPR strategies aggregate QoS classes (sort of DiffServ approach) simple set-up: 2 priority classes High-Priority (HP) traffic: always routed along the shortest path (SPR) using node-level resource partitioning limited packet loss limited delay and packet jitter Low-Priority (LP) traffic: two options always routed along the shortest path (SPR) using spare resources overflow traffic re-routed to alternative paths (MPR) 36 18

Simple test topology 1 2 0 4 3 Average node degree: E = 2.4 16 wavelengths per link, each loaded with 0.8 traffic matrix generated accordingly 37 QoS performance SPR/MPR is the routing policy adopted for LP traffic HP traffic (20%) always uses SPR Accurate dimensioning gives a good degree of traffic differentiation LP routing policy does not affect HP LP performance is slightly affected by HP resource dimensioning (within the range considered) Packet loss probability 1 1e-01 1e-02 1e-03 1e-04 1e-05 1e-06 SPR HP SPR LP MPR LP 1e-07 1e-08 MPR HP 1 2 3 4 No. of reserved wavelenghts (K) 38 19

QoS performance SPR/MPR is the routing policy adopted for LP traffic HP traffic always uses SPR K = 3 1 Packet loss probability 1e-01 1e-02 1e-03 1e-04 1e-05 1e-06 SPR LP SPR HP MPR HP MPR LP 1e-07 1e-08 10 20 30 40 50 Percentage of HP traffic 39 Reference topology 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Average node degree: E = 3.25 40 20

Topology 1: uniform matrix, balanced load 1.E00 LP loss HP loss 1.E-01 1.E-02 Loss rate 1.E-03 1.E-04 1.E-05 1.E-06 0 5 10 15 20 25 30 35 40 45 50 Link 41 Reference topology 2 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Average node degree: E = 5.75 42 21

Topology 2: uniform matrix, balanced load 1.E00 LP loss HP loss 1.E-01 1.E-02 Loss rate 1.E-03 1.E-04 1.E-05 1.E-06 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 Link 43 Topologies 1 & 2: no. of hops distribution 1.E+00 1.E-01 E = 3.25 E = 5.75 1.E-02 1.E-03 Distribution 1.E-04 1.E-05 1.E-06 1.E-07 1.E-08 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 No. of hops 44 22

Link failure recovery in OPS networks Connectionless approach Single link failure Link failure detection technique notified by physical layer time-based signaling-based During failure detection, loss of optical packets supposed to be transmitted on the failed link After failure detection, the recovery procedure is called and a MPR alternative path is used 45 Packet loss during failure detection Analytical model for the packet loss probability as a function of detection time p is the loss probability in a failure free scenario t f is the failure time d is the failure detection delay p t t f t f PL ( t) = 1 (1 p) t f < t < t f + d t t f 1 (1 p) t t f + d t f + d 46 23

Packet loss during failure detection 1 loss rate when d = d 3 loss rate when d = d 2 loss rate when d = d 1 Packet Loss Rate 0.1 0.01 loss rate without link failure failure time d 1 d 2 d 3 0.001 0.04 0.05 t f 0.06 0.07 0.08 0.09 0.1 Time 47 Adding resources for protection The MPR-based protection scheme needs additional resources to be effective First, the network is dimensioned to have a given average load per wavelength (e.g. 0.7) with relation to the input traffic matrix Then, further wavelengths are added to each fiber so that each node sees all its output fibers with the same capacity Additional cost due to protection is 73% of the initial cost in terms of number of wavelengths 48 24

Impact on throughput 3.55e+12 Network Throughput [bps] No protection resources 3.50e+12 3.45e+12 3.40e+12 3.35e+12 3.30e+12 3.25e+12 before failure during failure detection 0 0.02 0.04 0.06 0.08 0.1 0.12 Simulation Time [sec] after failure detection +73% protection resources 49 Out-of-sequence packet delivery A consequence of dynamic resource allocation techniques deployed to reduce congestion Implications: complex reordering operations at the optical network edges due to the huge bit rate of the optical channels throughput degradation at the transport/application level TCP congestion control highly affected by unordered segments real-time, UDP-based traffic requirements impaired by excessive delay due to unordered packets and/or reordering process Possible solutions: WDS policy with some time constraints in order to keep the correct packet sequence dynamic multi-path routing limited to delay-equivalent paths But what do we intend with correct packet sequence? 50 25

Alternatives for packet sequence ARRIVAL P n P n+1 t n t n+1 t DEPARTURE ❶ strictly in-sequence ❷ strictly in-sequence (transparent) ❸ strictly in-sequence ❹ loosely in-sequence ❺ loosely in- OR out-of-sequence ❻ ❼ L n+1 0 L n t n loosely out-of-sequence strictly out-of-sequence 51 s n Jitter distribution Jitter distribution G-type WDS with sequence constraint 10-1 1 10-2 10-3 10-4 10-5 10-6 1 2 3 4 5 6 7 Region Jitter distribution 10-1 1 10-2 10-3 10-4 10-5 10-6 G-type WDS 1 2 3 4 5 6 7 Region 52 26

Impact on loss 1 Packet loss probability 10-1 10-2 10-3 10-4 10-5 10-6 with sequence constraint without sequence constraint 10-7 0 0.5 1 1.5 2 2.5 D (normalized to the average packet length) 53 Conclusions OPS from the network perspective: issues dynamic multi-path routing for contention resolution QoS differentiation through resource partitioning and differentiated routing strategy link failure protection strategy through multi-path routing and additional resources provisioning packet sequence issues Further ongoing activities: effective low-complexity implementation of void filling WDS study of dynamic routing through ant colony optimization study of the impact of packet sequence break on higher layer protocols study of the impact of limited wavelength conversion on node and network performance 54 27