Wavelength Band Switching in Multigranular Optical WDM Networks

Transcription

1 Wavelegth Bad Switchig i Multigraular Optical WDM Networks Vishal Aad Collaborators X. Cao,, Dr. Y. Xiog ad Dr. C. Qiao LANDER, CSE Departmet, SUNY at Buffalo ~1~ Vishal Aad

2 Outlie λ The Problem with preset WDM etworks λ Cocept of wavelegth bad switchig: 3 layer wavebad switchig OXC architecture λ Wavelegth Bad Switchig: Schemes ad Groupig Strategies λ Wavelegth Bad Switchig Vs Wavelegth Routed Networks: How similar, How differet λ Performace of Wavelegth Bad Switchig: Techiques for Static ad Dyamic Traffic: ILP, Algorithms ad simulatio results λ A Sigle-layer wavebad switchig OXC architecture λ Wavelegth Vs Wavebad Coversio λ New techiques for failure recovery i WBS etworks λ Coclusios ~2~ Vishal Aad

3 Preset WDM etworks : The Problem λ Iteret traffic demad o the rise λ Oly way to keep up: WDM λ Causes deploymet of more fibers ad more wavelegths per fiber (DWDM) λ I-tur implies icreased size of Optical Cross-coects (OXC), with large port couts λ Hece, maagig/cotrollig (NMS, EMS) this large amout of traffic, associated resources: critical, difficult, complicated! ~3~ Vishal Aad

4 Preset WDM etworks : The Problem (cot'd) λ This traslates to icreased cost: both Capital (CAPEX) ad Operatig (OPEX) λ Despite the techological advaces: λ i WDM, Photoic-XC systems, switchig fabric λ The deploymet ad potetial use is limited λ Uprove reliability ad costs of huge switches (e.g. 1000x1000 ports) λ Large footprit (size), power requiremets ad (u) scalability cocers ~4~ Vishal Aad

5 The typical Optical Cross-Coect (OXC) λ Switchig at a optical ode too may wavelegth-ports ~5~ Vishal Aad

6 Wavelegth Bad Switchig λ Wavelegth bad: a group of several wavelegths λ WBS: A ew switchig hierarchy with multiple graularity λ WBS Networks: Use WBS i cojuctio with a multi-graular OXC, MG-OXC Switch each wavelegth idividually Typical-OXC Total ports = 4 + 4x2 + 4x2 + 4 = 24+add/drop+mux/demux A B D C λ 0 λ2 1 λ 3 MG-OXC b 0 A Switch bad of 4 wavelegths usig 1 port Total Ports = 1 + 1x2 + 1x2 + 1 = 6+add/drop+mux/demux oly! b 0 B C b 0 D b 0 ~6~ Vishal Aad

7 A Three-layer MG-OXC λ Ay fiber (bads) ca be demultiplexed ito bads (wavelegths) usig FTB/BTW λ Ay bad (wavelegth) ca be multiplexed ito fibers (bads) usig BTF/WTB λ Fibers/bads/wavelegths are switched at the FXC/BXC/WXC-layers λ Port types λ Cross-coect bypass traffic λ Add/drop add/drop traffic λ Mux/Demux muxed/demuxed traffic ~7~ Vishal Aad

8 Example:WBS usig a 3-layer MG-OXC λ 0 λ λ λ 2 idividual lightpaths: λ 0 o fiber F 1 bypassig the ode ad λ 1 to be added locally After demux. F 1, λ 0 is extracted ad grouped with λ 1 after goig thro muxer(s) Fially λ 0 ad λ 1 are mux. (combied) i a bad ad go out o fiber F 2 together λ 1 ~8~ Vishal Aad

9 Classificatio of WBS Schemes λ λ λ Use the pre-determied wavelegth set scheme as it is the simplest Each fiber has a fixed # of bads (B), each bad has a fixed umber of wavelegths (W), which are cosecutive (pre-determied) Or : each fiber has a fixed # B, each bad has a fixed # of wavelegths ad these wavelegths are chose radomly (ot ecessarily cosecutively) / adaptively may be more flexible, BUT too complex to realize i practice ~9~ Vishal Aad

10 Wavebad Groupig strategies (1) ed-to-ed: groupig the lightpaths with the same source-destiatio pair oly; (2) oe-ed:groupig the lightpaths from the same source oly OR groupig the lightpaths with same destiatio oly; (3) sub-path: groupig the lightpaths with commo itermediate liks (i.e. sub-paths); From ay source to ay destiatio λ Strategy (3) is the most geeral, BUT also complex ~10~ Vishal Aad

11 WBS Vs classical Wavelegth Routed Networks (WRN) λ Differet objectives ad techiques λ WRN: typically miimize wavelegths or wavelegth-hops(wh) λ WBS etworks: miimize the umber of ports λ Miimizig wavelegths does ot miimize the um. of ports λ Used a algorithm which optimizes (usig Liear Prog.) the used wavelegths by Routig & Wavelegth Assigmet (RWA), ad the does best effort groupig, backfired λ Caused a icrease rather tha a decrease i port cout λ A ideal WBS algo. may eed to trade a slight icrease i wavelegths for a much reduced port cout λ The WBS optimizatio problem has more costraits ad harder to solve ~11~ Vishal Aad

12 WBS Vs WRN (cot'd) λ Techiques developed for WRN, traffic groomig caot be applied directly to address WBS-problems λ I WRN, traffic groomig is used to reduce (de) mux, electroics, wavelegths ad hece cost λ WRN: ay set of lower bit rate sub-wavelegth traffic ca be multiplexed oto a wavelegth λ Oly costrait is total bit rate max. bit rate of wavelegth λ E.g. ay 12 SONET STS-1 (51.84 Mbps) sigals ca be multiplexed oto a OC-12 wavelegth, as: 12 x STS-1 = Mbps = OC-12 (2.5Gbps) λ WBS: at least oe more costrait λ Oly the traffic carried by a fixed set(typically cosecutive) ca be grouped ito a bad ~12~ Vishal Aad

13 Performace of WBS etworks: Static λ The problem: λ Give: λ case experimetal Results λ Network topology, umber of wavelegths per fiber (F), bads per fiber (B) ad graularity (W), etwork capacity is ot fixed λ A set of static traffic demads (i.e. lightpaths) all give upfrot How to satisfy all the traffic usig miimum umber of ports, whe o wavelegth coversio is available? λ Approach: 1. Optimizatio usig Iteger liear Programmig (ILP) model, for details Refer [opticomm 02, Ifocom 03] λ Not feasible (uses too much time ad memory) for large problem sizes 2. Heuristic based approach for large problems λ Based o wavebad assigmet strategy (3), sub-path groupig ~13~ Vishal Aad

14 ILP formulatio for WBS cetral ideas λ Objective: MIN MIN [ α (max WXC [ α + β WXC + γ α = β = γ = 1 all ports have equal cost + BXC β + γ BXC FXC ] 1( OR ) FXC ]) 2 λ Miimize the total umber of ports λ Miimize the maximum size of a ode (i.e. port cout) λ Variables λ Node (i.e. ports therei) is the cetral poit of iterest λ Defie the property or characteristics of a ode istead of a lik ~14~ Vishal Aad

15 ILP formulatio for WBS (cot'd) λ Costraits λ RWA: similar to traditioal RWA ILP λ Flow coservatio λ Wavelegth capacity λ Wavelegth cotiuity λ Wavebad switchig λ Bypass lightpath uses exactly oe of FXC/BXC/WXC crosscoect λ Add lightpath uses exactly oe add port at FXC/BXC/WXC λ Drop lightpath uses exactly oe drop port at FXC/BXC/WXC λ Mux/demux Every added wavelegth has to use a WTB mux (port) Every bad has to use a BTF mux (port) before leavig ode ~15~ Vishal Aad

16 ILP formulatio for WBS (cot'd) λ Port umbers at a ode : 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 bads ad ports from WTB ad BTW, mux/demux λ FXC layer: sum of ports for bypass/add/drop fibers ad ports from FTB ad BTF, mux/demux λ RWA ad groupig is doe so that the ILP miimizes the total umber of ports above ~16~ Vishal Aad

17 Heuristic Algorithms λ WBO-RWA:Wavebad Oblivious (but optimal) RWA λ Use ILP formulatios for traditioal RWA that miimize the total umber of used wavelegth-hop (WH) λ The group the assiged wavelegths ito bads ad calculate the umber of required ports λ Best effort groupig, doe as a afterthought, completely oblivious to the existece of wavebads λ BPHT: Balaced Path with Heavy-Traffic first wavebad assigmet λ Variatios of BPHT, e.g. BTMH:balaced traffic with max-hop first, BPMH:balaced path with max-hop first λ Performace of BPHT the best λ Hece show results of BPHT, WBO-RWA ad ILP ~17~ Vishal Aad

18 BPHT cetral ideas λ To maitai wavelegth-cotiuity (o coversio), assig loger paths, with more WHs first, to reduce blockig λ Assig bypass lightpaths (typically 60-80% of the traffic) first λ Assig paths that have maximum liks i commo to reduce ports by switchig them together as a bad λ Stage 1: Load Balaced K shortest path (KSP) Routig λ Start with the ode-pair with max. WHs alog its shortest path λ Use k-shortest paths for every ode-pair λ Load balace by miimizig the maximum load o a lik ~18~ Vishal Aad

19 BPHT cetral ideas (cot'd) λ λ Stage 2: Wavelegth Assigmet λ Cosider all ode-pair traffic with hops (h p 2) first λ Defie a set Q sd for every ode pair (s,d), which icludes all traffic whose start/ed are alog the path from s to d λ Calculate weight of each set Q sd,, W sd = hp * Tp p Q sd λ Startig with the largest weight set, util all traffic is satisfied 1. Assig wavelegths to all traffic from same source s 2. Assig wavelegths to all traffic from same destiatio d 3. Recursively assig the remaiig lightpaths i the set similarly λ Stage 3: Switchig Oce wavelegths are assiged, switch as may lightpaths usig fibers, the remaiig as bads, ad fially the still remaiig idividually at the wavelegth level ~19~ Vishal Aad

20 State, after load balaced routig S 0 S 1 S 4 Illustratio of BPHT S 2 S W S P 4 D 2 D 0 D 1 LPs= {4,5,6 } S 4 D 2 = hp t p p = 9 S 0 WS 1= = = 14 0 D hp tp x x x x p S P 0 D 1 S D 2 LPs={1,2,3,4} S 1 S 2 S 3 b 1 λ 2 b 2 λ 3 λ 4 D 2 λ 0 λ 1 D 0 D 1 λ 5 b 0 W S 4 D 2 = hp t p S P 4 D 2 LPs= {5,6} p = 7 ~20~ Vishal Aad

21 Performace Evaluatio λ Defie 3 Performace Metrics: each metric is a fuctio of the WBS algorithm a 1. Total port umber ratio T(a) Total( FXC + BXC Total( OXC + WXC ) u si g WBS a lg orithm ) of ordiary OXC ' a' 2. Max port umber ratio M(a) Max( FXC + BXC Max( OXC + WXC ) u si g WBS a lg orithm ' a' ) of ordiary OXC 3. Used wavelegth chaels ratio W(a) λ hop used by WBS a lg orithm ' a' λ hop used by optimal RWA without WBS λ Improvemet i port cout is: 1-T(a) λ By defiitio W(WBO-RWA) = 1 ~21~ Vishal Aad

22 Simulatio Results I λ Results of ILP model for a small etwork ~22~ Vishal Aad

23 Simulatio Results II λ Results for a large etwork Radom traffic ~23~ Vishal Aad

24 Simulatio Results III λ Results for a large etwork Uiform traffic, W=4 ~24~ Vishal Aad

25 Performace of WBS etworks: Dyamic case experimetal Results λ λ The problem: λ Give: λ λ λ Network topology, with fixed umber of wavelegths per fiber (F), bads per fiber (B) ad graularity (W), ow etwork capacity is fixed/limited A set of dyamic icremetal traffic demads (i.e. lightpaths): demads arrive oe after the other, with o kowledge of future demads How to satisfy maximum traffic (i.e. with low blockig) usig miimum umber of ports, whe o wavelegth coversio is available? Approach: 1. Limited recofiguratio (to save ports) MG-OXC architecture 2. Heuristic MILB: Maximum Iterferece Legth I Bad λ Based o assigig lightpaths routes & wavelegths i a bad such that the umber of liks shared with existig lightpaths i that bad is maximized ~25~ Vishal Aad

26 Dyamically recofigured MG-OXC λ To save o ports, do ot allow full recofiguratio λ Istead of allowig ay fiber/ bad to be demuxed, allow oly a limited umber of bads to be demuxed λ Num. bads i a fiber = Y, allow oly βy bads to be demuxed ito wavelegths, β<1 λ Hece oly limited recofigurability allowed ~26~ Vishal Aad

27 MILB for Dyamic icremetal traffic λ Model the etwork as a bad-graph with B layers oe for each bad S 1 S 2 S 3 λ 0 S 0 S 1 S 2 S 3 k 0 S 4 S 5 S 6 S 7 S 0 b 0 S 4 S 5 S 6 S 7 Bad layer -1 existig lightpaths 2 possible routes k 1 S 0 S 1 S 2 S 3 b 1 S 4 S 5 S 6 S 7 ew lightpath λ 3 λ 2 Bad layer -2 ~27~ Vishal Aad

28 Performace Results: Dyamic Case Compared MILB with First-Fit(FF) ad Radom -fit (RF) for various bad sizes ad β λ MILB performs the best λ Switches max. lightpaths as a group (bad) λ FF ad RF oly group as a afterthought, For same umber of ports RF, FF block more lightpaths λ β=0.44, MILB achieves least blockig with least port cout λ Blockig does ot decrease with icrease i β, oly due to lack of wavelegths, ot ports λ Savigs = (1-0.44) 60% λ Oly at β = 1, FF has lower blockig, BUT at the expese of large port cout! λ Suggests havig/buildig-i 44% BTW ports, BUT ot more ~28~ Vishal Aad

29 Wavelegth Hop Vs Num. Port λ Trade-off : Wavelegth Hop (WH) Vs Num. of Ports λ While usig ILP ad heuristics for the Static Case ad heuristics for Dyamic Case λ WHY? λ Do ot always use the shortest WH path λ May use a loger WH path which icreases used wavelegth resources but decrease ports λ Static Case: λ ILP: aturally chooses paths ad wavelegths which miimizes ports, WH miimizatio is secodary (a byproduct) λ BPHT: a loger path may be used i step 1, for load balacig λ Dyamic Case: λ MILB: may choose a loger path ad bad which has more iterferece ad maximizes badig ad port reductio, but this icreases WH λ Our techiques: perform WBS with oly a small icrease i wavelegth resources, with a large decrease i port cout ~29~ Vishal Aad

30 A Sigle-layer MG-OXC λ Oly certai desigated fibers (bads) ca be multiplexed or demultiplexed λ For example, i the fig. oly fiber ca be (de) multiplexed ito bads λ oly certai bads from this ca be (de) multiplexed ito wavelegths λ Fibers 1,2 simply pass-thro, switched at the FXC-layer ~30~ Vishal Aad

31 Sigle Vs Multi-layer MG-OXC Compariso criteria Sigle Layer Multi-layer Num. Of Switches Num. Of Mux/DeMux Cotrol Complexity Num. Of ports WBS algorithm Logical divisio: oly 1 switch,fabric: better sigal quality Elimiate FTB/BTW, BTF/WTB (de) mux s b/ layers Simpler architecture to implemet/cofigure/cotrol Ideally very few (fewer tha multi-layer) λvery Complex λpractically impossible to achieve port-cout eve close to the ideal case λresult i more blockig for dyamic traffic Physically 3 diff. Layers/switches/fabrics FTB/BTW, BTF/WTB (de)mux s still eeded Relatively more complex Practically few λrelatively simpler λpossible to achieve a small port-cout with a simpler algo. for both static ad dyamic traffic ~31~ Vishal Aad

32 Wavelegth ad Wavebad coversio λ Wavebad coversio is similar, but ot idetical to limited wavelegth coversio λ With bad coversio: covertig b 0 to b 1, causes ot oly lambda-0 lambda-1, but also forces the coversio of lambda-1 lambda-3 simultaeously λ I WRN, with full coversio, wavelegth assigmet is trivial λ I WBS, coversio does facilitate groupig ad ease wavelegth requiremet λ But wavelegth coversio does require a fiber/bad to be first demultiplexed ito wavelegths Potetially icreasig the port cout λ Wavebad coversio allows coversio without this demultiplexig ad additioal ports, but with the above costraits ~32~ Vishal Aad

33 New techiques for failure recovery λ New failures i WBS etworks: port, mux/demux/badcoverters, causig wavelegth ad bads to fail λ Thus, eve if a lik/fiber does ot fail (as i WRN), the above itra-fiber failures have to be accouted for λ Use ew Bad-Mergig ad Bad-Swappig methods ~33~ Vishal Aad

34 Coclusios λ Itroduced the cocept of Wavelegth Bad Switchig (WBS) λ Explored the advatage of WBS, ad developed itelliget WBS algorithms, optical cross-coect architectures λ Developed ILP formulatios ad heuristics to cosider the efficiet desig of WBS optical etworks for both Static ad Dyamic traffic λ Itelliget WBS algorithms heuristics (such as BPHT, MILB) ca save cosiderably o port cout. Bad heuristics such as WBO- RWA may eed eve more ports tha ordiary-oxc etwork λ There is a trade-off betwee wavelegth-hop used ad the total port cout i MG-OXC etwork λ WBS to reduce ports requires oly a small icrease i wavelegth resources for a much larger decrease i port cout (etwork cost) λ Cosidered critical WBS issues such as coversio ad failure recovery ~34~ Vishal Aad