Wavelength Band Switching in Multigranular Optical WDM Networks Vishal Anand Collaborators X. Cao, Dr. Y. Xiong and Dr. C. Qiao LANDER, CSE Department, SUNY at Buffalo ~1~ Vishal Anand Outline λ The Problem with present WDM networks λ Concept of wavelength band switching: 3 layer waveband switching OXC architecture λ Wavelength Band Switching: Schemes and Grouping Strategies λ Wavelength Band Switching Vs Wavelength Routed Networks: How similar, How different λ Performance of Wavelength Band Switching: Techniques for Static and Dynamic Traffic: ILP, Algorithms and simulation results λ A Single-layer waveband switching OXC architecture λ Wavelength Vs Waveband Conversion λ New techniques for failure recovery in WBS networks λ Conclusions ~2~ Vishal Anand Present WDM networks : The Problem λ Internet traffic demand on the rise λ Only way to keep up: WDM λ Causes deployment of more fibers and more wavelengths per fiber (DWDM) λ In-turn implies increased size of Optical Cross-connects (OXC), with large port counts λ Hence, managing/controlling (NMS, EMS) this large amount of traffic, associated resources: critical, difficult, complicated! ~3~ Vishal Anand Present WDM networks : The Problem (cont'd) λ This translates to increased cost: both Capital (CAPEX) and Operating (OPEX) λ Despite the technological advances: λ in WDM, Photonic-XC systems, switching fabric λ The deployment and potential use is limited Unproven reliability and costs of huge switches (e.g. 1000x1000 ports) λ Large footprint (size), power requirements and (un) scalability concerns λ ~4~ Vishal Anand The typical Optical Cross-Connect (OXC) λ Switching at an optical node—too many wavelength-ports ~5~ Vishal Anand Wavelength Band Switching λ Wavelength band: a group of several wavelengths λ WBS: A new switching hierarchy with multiple granularity λ WBS Networks: Use WBS in conjunction with a multi-granular OXC, MG-OXC Typical-OXC A λ λ01 λ2 λ3 MG-OXC Switch each wavelength individually Total ports = 4 + 4x2 + 4x2 + 4 = 24+add/drop+mux/demux D B C Switch band of 4 wavelengths using 1 port Total Ports = 1 + 1x2 + 1x2 + 1 = 6+add/drop+mux/demux only! B A b0 C b0 D b0 b0 ~6~ Vishal Anand A Three-layer MG-OXC λ Any fiber (bands) can be demultiplexed into bands (wavelengths) using FTB/BTW λ Any band (wavelength) can be multiplexed into fibers (bands) using BTF/WTB λ Fibers/bands/wavelengths are switched at the FXC/BXC/WXC-layers λ Port types λ Cross-connect – bypass traffic λ Add/drop – add/drop traffic λ Mux/Demux – muxed/demuxed traffic ~7~ Vishal Anand Example:WBS using a 3-layer MG-OXC λ0 λ 2 individual lightpaths: λ0 on fiber F1 bypassing the node and λ1 to be added locally λ After demux. F1, λ0 is extracted and grouped with λ1 after going thro muxer(s) λ Finally λ0 and λ1 are mux. (combined) in a band and go out on fiber F2 “together” λ1 ~8~ Vishal Anand Classification of WBS Schemes λ λ λ Use the pre-determined wavelength set scheme as it is the simplest Each fiber has a fixed # of bands (B), each band has a fixed number of wavelengths (W), which are consecutive (pre-determined) Or : each fiber has a fixed # B, each band has a fixed # of wavelengths and these wavelengths are chosen randomly (not necessarily consecutively) / adaptively — may be more flexible, BUT too complex to realize in practice ~9~ Vishal Anand Waveband Grouping strategies (1) end-to-end: grouping the lightpaths with the same source-destination pair only; (2) one-end:grouping the lightpaths from the same source only OR grouping the lightpaths with same destination only; (3) sub-path: grouping the lightpaths with common intermediate links (i.e. sub-paths); From any source to any destination λ Strategy (3) is the most general, BUT also complex ~10~ Vishal Anand WBS Vs classical Wavelength Routed Networks (WRN) λ Different objectives and techniques λ WRN: typically minimize wavelengths or wavelength-hops(WH) λ WBS networks: minimize the number of ports λ Minimizing wavelengths does not minimize the num. of ports λ Used an algorithm which optimizes (using Linear Prog.) the used wavelengths by Routing & Wavelength Assignment (RWA), and then does best effort grouping, backfired λ Caused an increase rather than a decrease in port count λ An ideal WBS algo. may need to trade a slight increase in wavelengths for a much reduced port count λ The WBS optimization problem has more constraints and harder to solve ~11~ Vishal Anand WBS Vs WRN (cont'd) λ Techniques developed for WRN, traffic grooming cannot be applied directly to address WBS-problems λ In WRN, traffic grooming is used to reduce (de) mux, electronics, wavelengths and hence cost λ WRN: any set of lower bit rate sub-wavelength traffic can be multiplexed onto a wavelength λ Only constraint is total bit rate ≤ max. bit rate of wavelength λ E.g. any 12 SONET STS-1 (51.84 Mbps) signals can be multiplexed onto a OC-12 wavelength, as: 12 x STS-1 = 622.08 Mbps = OC-12 (2.5Gbps) λ WBS: at least one more constraint λ Only the traffic carried by a fixed set (typically consecutive) can be grouped into a band ~12~ Vishal Anand Performance of WBS networks: Static case experimental Results λ The problem: λ Given: λ Network topology, number of wavelengths per fiber (F), bands per fiber (B) and granularity (W), network capacity is not fixed λ A set of static traffic demands (i.e. lightpaths) – all given upfront λ How to satisfy all the traffic using minimum number of ports, when no wavelength conversion is available? λ Approach: 1. Optimization using Integer linear Programming (ILP) model, for details Refer [opticomm’02, Infocom’03] λ Not feasible (uses too much time and memory) for large problem sizes 2. Heuristic based approach for large problems λ Based on waveband assignment strategy (3), sub-path grouping ~13~ Vishal Anand ILP formulation for WBS — central ideas λ Objective: MIN [α ∑ WXC n + β ∑ BXC n + γ ∑ FXC n ] → 1(OR ) n n n MIN ( max [α ∑ WXC n + β ∑ BXC n + γ ∑ FXC n ]) → 2 n n n n α = β = γ = 1 ⇒ all ports have equal cost λ Minimize the total number of ports λ Minimize the maximum size of a node (i.e. port count) λ Variables λ Node (i.e. ports therein) is the central point of interest λ Define the property or characteristics of a node instead of a link ~14~ Vishal Anand ILP formulation for WBS (cont'd) λ Constraints λ RWA: similar to traditional RWA ILP λ Flow conservation λ Wavelength capacity λ Wavelength continuity λ Waveband switching λ Bypass lightpath uses exactly one of FXC/BXC/WXC crossconnect λ Add lightpath uses exactly one add port at FXC/BXC/WXC λ Drop lightpath uses exactly one drop port at FXC/BXC/WXC λ Mux/demux – Every added wavelength has to use a WTB mux (port) – Every band has to use a BTF mux (port) before leaving node n ~15~ Vishal Anand ILP formulation for WBS (cont'd) λ Port numbers at a node : calculated from the values of the variables λ WXC layer: sum of bypass/add/drop lightpaths λ BXC layer: sum of ports for bypass/add/drop bands and ports from WTB and BTW, mux/demux λ FXC layer: sum of ports for bypass/add/drop fibers and ports from FTB and BTF, mux/demux λ RWA and grouping is done so that the ILP minimizes the total number of ports above ~16~ Vishal Anand Heuristic Algorithms λ WBO-RWA:Waveband Oblivious (but optimal) RWA λ Use ILP formulations for traditional RWA that minimize the total number of used wavelength-hop (WH) λ Then group the assigned wavelengths into bands and calculate the number of required ports λ Best effort grouping, done as an afterthought, completely oblivious to the existence of wavebands λ BPHT: Balanced Path with Heavy-Traffic first waveband assignment λ Variations of BPHT, e.g. BTMH:balanced traffic with max-hop first, BPMH:balanced path with max-hop first λ Performance of BPHT the best λ Hence show results of BPHT, WBO-RWA and ILP ~17~ Vishal Anand BPHT– central ideas λ To maintain wavelength-continuity (no conversion), assign longer paths, with more WHs first, to reduce blocking λ Assign bypass lightpaths (typically 60-80% of the traffic) first λ Assign paths that have maximum links in common to reduce ports by switching them together as a band λ Stage 1: Load Balanced K shortest path (KSP) Routing λ Start with the node-pair with max. WHs along its shortest path λ Use k-shortest paths for every node-pair λ Load balance by minimizing the maximum load on a link ~18~ Vishal Anand BPHT– central ideas (cont'd) λ Stage 2: Wavelength Assignment λ λ λ λ λ Consider all node-pair traffic with hops (hp ≥ 2) first Define a set Qsd for every node pair (s,d), which includes all traffic whose start/end are along the path from s to d Calculate weight of each set Qsd,, Wsd = ∑ h p * Tp p∈Q Starting with the largest weight set, until all traffic is satisfied 1. Assign wavelengths to all traffic from same source ‘s’ 2. Assign wavelengths to all traffic from same destination ‘d’ 3. Recursively assign the remaining lightpaths in the set similarly sd Stage 3: Switching λ Once wavelengths are assigned, switch as many lightpaths using fibers, the remaining as bands, and finally the still remaining individually at the wavelength level ~19~ Vishal Anand Illustration of BPHT State, after load balanced routing WS4D2 = ∑hp×tp =9 S4 S0 S1 S2 S3 2 1 D0 D1 3 4 6 5 LPs={1,2,3,4} 1 S4 S0 S1 2 LPs ={4,5,6} D2 WS0D1 = ∑hp×tp =5x1+4x1+3x1+2x1=14 p∈PDS0 p∈PDS4 WS4D2 = ∑hp×tp =7 p∈PDS4 2 b0 S2 S3 b2 b1 D0 λ3 λ1 λ0 LPs={5,6} D1 λ2 λ4 λ5 D2 ~20~ Vishal Anand Performance Evaluation λ Define 3 Performance Metrics: each metric is a function of the WBS algorithm ‘a’ 1. Total port number ratio T(a) Total ( FXCn + BXCn + WXCn )u sin g WBS a lg orithm ' a ' Total (OXCn ) of ordinary − OXC 2. Max port number ratio M(a) Max( FXCn + BXCn + WXCn )u sin g WBS a lg orithm ' a ' Max(OXCn ) of ordinary − OXC 3. Used wavelength channels ratio W(a) λ − hop used by WBS a lg orithm ' a' λ − hop used by optimal RWA without WBS λ λ Improvement in port count is: 1-T(a) By definition W(WBO-RWA) = 1 ~21~ Vishal Anand Simulation Results I λ Results of ILP model for a small network ~22~ Vishal Anand Simulation Results II λ Results for a large network—Random traffic ~23~ Vishal Anand Simulation Results III λ Results for a large network—Uniform traffic, W=4 ~24~ Vishal Anand Performance of WBS networks: Dynamic case experimental Results λ The problem: λ Given: λ λ Network topology, with fixed number of wavelengths per fiber (F), bands per fiber (B) and granularity (W), now network capacity is fixed/limited A set of dynamic incremental traffic demands (i.e. lightpaths): demands arrive one after the other, with no knowledge of future demands λ How to satisfy maximum traffic (i.e. with low blocking) using minimum number of ports, when no wavelength conversion is available? λ Approach: 1. 2. Limited reconfiguration (to save ports) MG-OXC architecture Heuristic MILB: Maximum Interference Length In Band λ Based on assigning lightpaths routes & wavelengths in a band such that the number of links shared with existing lightpaths in that band is maximized ~25~ Vishal Anand Dynamically reconfigured MG-OXC λ To save on ports, do not allow full reconfiguration λ Instead of allowing any fiber/ band to be demuxed, allow only a limited number of bands to be demuxed λ Num. bands in a fiber = Y, allow only βY bands to be demuxed into wavelengths, β<1 λ Hence only limited reconfigurability allowed ~26~ Vishal Anand MILB for Dynamic incremental traffic λ Model the network as a band-graph with B layers – one for each band λ0 S1 S2 S0 S4 k0 S1 S2 S3 S7 S0 S4 S5 2 possible routes S6 b0 S5 S3 S7 S6 Band layer -1 existing lightpaths S1 k1 S2 S0 new lightpath λ3 S4 λ2 S5 S3 b1 S7 S6 Band layer -2 ~27~ Vishal Anand Performance Results: Dynamic Case λ MILB performs the best λ Switches max. lightpaths as a group (band) λ FF and RF only group as an afterthought, For same number of ports RF, FF block more lightpaths λ β = 0.44, MILB achieves least blocking with least port count λ Blocking does not decrease with increase in β, only due to lack of wavelengths, not ports λ Savings = (1- 0.44) ≅ 60% λ Only at β = 1, FF has lower blocking, BUT Compared MILB with First-Fit(FF) at the expense of large port count ! and Random -fit (RF) for various λ Suggests having/building-in 44% BTW band sizes and β ports, BUT not more ~28~ Vishal Anand Wavelength Hop Vs Num. Port λ Trade-off : Wavelength Hop (WH) Vs Num. of Ports λ While using ILP and heuristics for the Static Case and heuristics for Dynamic Case λ WHY? λ Do not always use the shortest WH path λ May use a longer WH path which increases used wavelength resources but decrease ports λ Static Case: λ ILP: naturally chooses paths and wavelengths which minimizes ports, WH minimization is secondary (a byproduct) λ BPHT: a longer path may be used in step 1, for load balancing λ Dynamic Case: λ MILB: may choose a longer path and band which has more interference and maximizes banding and port reduction, but this increases WH λ Our techniques: perform WBS with only a small increase in wavelength resources, with a large decrease in port count ~29~ Vishal Anand A Single-layer MG-OXC λ Only certain “designated” fibers (bands) can be multiplexed or demultiplexed λ For example, in the fig. only fiber ‘n’ can be (de) multiplexed into bands λ only certain bands from this can be (de) multiplexed into wavelengths λ Fibers 1,2 simply pass-thro, switched at the FXC-layer ~30~ Vishal Anand Single Vs Multi-layer MG-OXC Comparison criteria Single Layer Multi-layer Num. Of Switches Logical division: only 1 switch,fabric: better signal quality Physically 3 diff. Layers/switches/fabrics Num. Of Mux/DeMux Eliminate FTB/BTW, BTF/WTB (de) mux’s b/n layers FTB/BTW, BTF/WTB (de)mux’s still needed Control Complexity Simpler architecture to Relatively more complex implement/configure/control Num. Of ports Ideally very few (fewer than multi-layer) Practically few WBS algorithm λVery Complex λPractically impossible to achieve port-count even close to the ideal case λResult in more blocking for dynamic traffic λRelatively simpler λPossible to achieve a small port-count with a simpler algo. for both static and dynamic traffic ~31~ Vishal Anand Wavelength and Waveband conversion λ Waveband conversion is similar, but not identical to limited wavelength conversion λ With band conversion: converting b0 to b1, causes not only lambda-0 Ælambda-1, but also forces the conversion of lambda-1 Ælambda-3 simultaneously λ In WRN, with full conversion, wavelength assignment is trivial λ In WBS, conversion does facilitate grouping and ease wavelength requirement λ But wavelength conversion does require a fiber/band to be first demultiplexed into wavelengths Potentially increasing the port count λ Waveband conversion allows conversion without this demultiplexing and additional ports, but with the above constraints ~32~ Vishal Anand New techniques for failure recovery λ New failures in WBS networks: port, mux/demux/bandconverters, causing wavelength and bands to fail λ Thus, even if a link/fiber does not fail (as in WRN), the above intra-fiber failures have to be accounted for λ Use new Band-Merging and Band-Swapping methods ~33~ Vishal Anand Conclusions λ Introduced the concept of Wavelength Band Switching (WBS) λ Explored the advantage of WBS, and developed intelligent WBS algorithms, optical cross-connect architectures λ Developed ILP formulations and heuristics to consider the efficient design of WBS optical networks for both Static and Dynamic traffic λ Intelligent WBS algorithms heuristics (such as BPHT, MILB) can save considerably on port count. Bad heuristics such as WBORWA may need even more ports than ordinary-OXC network λ There is a trade-off between wavelength-hop used and the total port count in MG-OXC network λ WBS to reduce ports requires only a small increase in wavelength resources for a much larger decrease in port count (network cost) λ Considered critical WBS issues such as conversion and failure recovery ~34~ Vishal Anand