1 / 42

WP2 UPC Contribution to A2.2.1: Route Management

WP2 UPC Contribution to A2.2.1: Route Management. Route management. Network models Routing management in WS/OBS/OPS networks Routing models / Route management models Combined intra- and inter domain routing model(s) [ETH, UPC] Adaptive routing model(s) Predictive routing model(s)

larya
Télécharger la présentation

WP2 UPC Contribution to A2.2.1: Route Management

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. WP2UPC Contribution to A2.2.1: Route Management NOBEL WP2: Valladolid Oct. 13-15, 2004

  2. Route management • Network models • Routing management in WS/OBS/OPS networks • Routing models / Route management models • Combined intra- and inter domain routing model(s) [ETH, UPC] • Adaptive routing model(s) • Predictive routing model(s) • Multicast routing model NOBEL WP2: Valladolid Oct. 13-15, 2004

  3. Routing Management in WS/OBS/OPS Networks • Routing in WS → Routing and Wavelength Assignment (RWA) • Routing Problem • SPF • Alternate (k- Shortest Path, k- disjoint Shortest Path) • Adaptive (Pre-computed) • Wavelength Assignment • Random • Complex Algorithm NOBEL WP2: Valladolid Oct. 13-15, 2004

  4. Routing Management in WS/OBS/OPS Networks • Routing in OBS → Off-set between Burst Control Packet/data Burst • Routing • Source-based Connection-Oriented • BYPASS Based Routing (BBR) • Distributed packet-adaptive path performed on a set of pre-selected path • QoS • Low Priority (LP) burst • Multipath Source Routing with Shortest Path Selection • Distributed Adaptive Routing • High Priority (HP) burst • Adaptive routing • Another solution → BBR • HP bypass OSL • LP travel over OSL • Routing in OPS • Distributed packet-adaptive routing → inaccuracy ↓, complexity ↑ • Source Adaptive routing → BBR NOBEL WP2: Valladolid Oct. 13-15, 2004

  5. Combined intra- and inter domain routing models • Objectives • To provide a highly efficient coupling between both routing protocols with the aim that the combined QoSR model could be able to supply multi-constrained end-to-end optical paths closer to optimality • To provide a routing strategy, where all possible routing information can be used in the routing decision in order to find an optimal end-to-end path NOBEL WP2: Valladolid Oct. 13-15, 2004

  6. Combined intra- and inter domain routing models • Optical Network Models NOBEL WP2: Valladolid Oct. 13-15, 2004

  7. Combined intra- and inter domain routing models • Combined Routing Models to Exchange Routing Information Hybrid Model for the core Optical Network based on Inter-Domain Routing Agents (IDRA) NOBEL WP2: Valladolid Oct. 13-15, 2004

  8. Combined intra- and inter domain routing models • These IDRAs are responsible then for carrying inter-domain routing information, and deciding within each NSP which path is the best, among the available paths to reach any known destination • IDRAs basically exchange the set of offered services, in addition to the combined cost to reach any destination known • Synthesizing the information exchanged between the IDRAs is mandatory in order to design a highly scalable QoSR model • This QoS cost metric will be an input to the routing algorithm running on each IDRA, and will be used by the NSPs to select the best path for any given destination. NOBEL WP2: Valladolid Oct. 13-15, 2004

  9. Combined intra- and inter domain routing models • Inter-Domain Routing advertisements flow between a destination domain and a source domain in our combined Intra-Domain and Inter-Domain QoSR model NOBEL WP2: Valladolid Oct. 13-15, 2004

  10. BYPASS Based Routing Solution • Key aspects in the BBR process are: • Decide which links should be bypassed • How to compute the bypass-paths • When bypass-paths should be used • The BBR mechanism consists of three steps: • Identify the Obstruct-Sensitive Links • Consider these OSLs when selecting the path • Compute the bypass-paths NOBEL WP2: Valladolid Oct. 13-15, 2004

  11. b a-b a λ1,λ3 λ1, λ2, λ3 λ1,λ3 λ1,λ3 λ1, λ2, λ3 λ1, λ2, λ3 λ1, λ2, λ3 λ1, λ2, λ3 1 1 1 λ1, λ2 λ1, λ2 λ1, λ2 λ1, λ2, λ3 2 λ1, λ2, λ3 λ1, λ2, λ3 λ1, λ2, λ3 λ1,λ3 λ1, λ2, λ3 λ1, λ2 λ1, λ2 λ1, λ2, λ3 λ1, λ3 λ1, λ3 λ1,λ2 λ1,λ2, λ3 d λ1 λ1 λ1 λ1, λ2, λ3 3 Path c c-d Blocked Path Routing Inaccuracy in WDM Networks • Routing inaccuracy effects in WS networks a, 1, 2, b, (λ1) c, 1, 2, d, (λ1) NOBEL WP2: Valladolid Oct. 13-15, 2004

  12. Bypass Based Optical Routing (BBOR) Solution • Apply the BBOR to WS networks implies to define: • A triggering policy • A routing mechanism • Our triggering policy: • Based on a threshold, N • Update messages sent after N changes • Our routing mechanism consists of three steps: • Identify the Obstruct-Sensitive Wavelengths (OSWs) • Consider these OSWs in the RWA problem • Compute optical bypass-paths NOBEL WP2: Valladolid Oct. 13-15, 2004

  13. BBOR Solution: Identify the OSWs (step 1) • Obstruct-Sensitive wavelength (OSW): • Due to the inaccuracy potentially a wavelength might not be available in a certain link at the path set-up time • This wavelength is defined as OSW and has to be bypassed • OSW formal definition: • A λi is OSW (λios ) on a certain link if R ≤ Tp where: • B: total number of a certain λi on a link • N: threshold value • R: current number of available λi on this link • Tp: threshold percentage → Percentage of N to decide when a λ is OSW NOBEL WP2: Valladolid Oct. 13-15, 2004

  14. BBOR Solution:Consider the OSWs in the RWA problem(step 2) Degree of obstruction • New parameter OSWi (L,F): • i: colour • L: number of links where λi is OSW • F: minimum number of λi‘s available along the selected lightpath • Two routing algorithms: • ALG1 • Prioritizes to minimize the obstruction • ALG2 • Prioritizes to minimize the congestion Degree of congestion NOBEL WP2: Valladolid Oct. 13-15, 2004

  15. A New Prediction-based Routingand Wavelength Assignment Mechanism for Optical Transport Networks NOBEL WP2: Valladolid Oct. 13-15, 2004

  16. Usual Routing Algorithms need messages to update TEDs with information about the network state. Network state information is not accurate: - Aggregating information. - Triggering of update messages. - Latency associated in flooding of the update messages. Routing Algorithms utilise inaccurate state information: Routing Inaccuracy Problem (RIP). Prediction Based Routing NOBEL WP2: Valladolid Oct. 13-15, 2004

  17. Prediction Based Routing Idea • Source nodes can learn which is the best path and wavelength without update messages. • Dynamic learning according to the routing information obtained in previous connection set-ups. (Based on branch prediction). NOBEL WP2: Valladolid Oct. 13-15, 2004

  18. Prediction Based RoutingNetwork state registration • In Branch Prediction the prediction if a branch instruction will be taken or not is done according to the history of previous outcomes of the branch (not knowing exactly the processor state). • Analogy with branch prediction: It is necessary to register the history of previous connections set-ups. • Register of previous set-ups for every ligth-path Wavelength Registers (WR). NOBEL WP2: Valladolid Oct. 13-15, 2004

  19. Prediction Based RoutingCycle • Cycle: Basic unit of time where the history state registers (WRs) are modified. • One Cycle: Part of the holding time, in our simulations 1 Cycle = 1/10 holding time. • Time measured in cycles • Holding time • Arrival time • Time between updates ( for First-Fit) NOBEL WP2: Valladolid Oct. 13-15, 2004

  20. 0 1 0 0 0 1 1 1 1 1 0 0 Prediction Based Routing Wavelength Registers There is one WR for every wavelength on a path (ligth-path) for every source-destination pair of nodes. Every cycle the WR is filled with a ‘0’ if a traffic request is being sent with this wavelength on that path in that cycle from the source node to the destination node and with a ‘1’ if it is not. Example of WR: the value on the right is the newest and the value on the left is the oldest NOBEL WP2: Valladolid Oct. 13-15, 2004

  21. Prediction Based Routing Prediction Tables • There is one Prediction Table (PT) for every wavelength on a path (ligth-path) for every source-destination pair of nodes. • Index to access PT obtained from the corresponding WR. • Prediction: Read two-bit counter value<2 not blocked, value>1 blocked NOBEL WP2: Valladolid Oct. 13-15, 2004

  22. Prediction Based Routing Algorithm • The PBR algorithm checks the counter value of the PT and the availability of the corresponding outgoing link towards that destination. Two shortest paths, SP1, SP2 and W wavelengths. NOBEL WP2: Valladolid Oct. 13-15, 2004

  23. Prediction Based Routing Algorithm • Special case: • If the PBR algorithm predicts that all wavelengths of SP1 and SP2 are unavailable, it tries to assign route and wavelength based only on the availability information of the outgoing links. • Updating the PT: • PT of the selected wavelength and path is updated by increasing the counter if connection request is blocked and decreasing if it is not. NOBEL WP2: Valladolid Oct. 13-15, 2004

  24. Prediction Based RoutingEvaluation • Simulations: Comparison between PBR and First-Fit. • Topology test: 15 nodes (2 sources, 2 destinations) with one-fibre links. • Connections arrivals (60,000) modeled by Poisson, each one requiring a full wavelength. • Number of wavelengths variable: 2, 3, 4 and 5 • Number of cycles between update messages for First-Fit variable :1, 5, 10, 20, 40. NOBEL WP2: Valladolid Oct. 13-15, 2004

  25. 12000 10616 10000 First-Fit lambda=2 First-Fit lambda=3 First-Fit lambda=4 First-Fit lambda=5 PBR lambda=2 PBR lambda=3 8000 PBR lambda=4 PBR lambda=5 Blocked Requests 6000 4000 3796 2000 287 57 0 0 5 10 15 20 25 30 35 40 45 Cycles N (update) Prediction Based Routing Results NOBEL WP2: Valladolid Oct. 13-15, 2004

  26. Prediction Based RoutingResults Evaluation • For 2 and 3 wavelengths First-Fit behaves better than PBR only when the update messages are every cycle (N=1) • From lambda=4 the PBR always behaves better than the First-Fit If there are enough wavelengths the PBR assigns better the routes and wavelengths. • When N=1 and with enough lambdas First-Fit assigns an occupied wavelength because two connections are requested at the two source nodes at the same time. One node assigns with outdated information.  PBR more learning capability than First-Fit NOBEL WP2: Valladolid Oct. 13-15, 2004

  27. Prediction Based Routing Results: Learning Capability Evolution of blocked requests, every new 100 requests, from 0 to 2000 requests for lambda=4 and N=1. NOBEL WP2: Valladolid Oct. 13-15, 2004

  28. Multicast-like approach for optical networks resources optimisation NOBEL WP2: Valladolid Oct. 13-15, 2004

  29. Outline • Introduction • Multicast-like approach to optimise the optical resources • Simulation case study • Multicast approach vs. Unicast approach • Conclusions • Future work NOBEL WP2: Valladolid Oct. 13-15, 2004

  30. Introduction • Optical channels are currently able to carry 10 and even 40 Gbit/s traffic streams • It is not usual to have such amounts of traffic between any pair of client nodes • Wavelengths capacity is wasted because there is a granularity mismatch between client data flows and optical channels capacity • Sublambdas concept: each clients connection request is for a fraction of the wavelength bandwidth • Different calls between the node-pair (i,j) can be allocated/groomed on the same wavelength • There is the need to optimise the optical resource utilisation while providing cost savings (reducing the optical transmission equipments) NOBEL WP2: Valladolid Oct. 13-15, 2004

  31. Multicast approach (I) • Unlike point-to-point optical connections (Unicast), every new optical connection established is routed not only to the destination node but also to (N) diverse nodes • The source node has more than one logical neighbour • The traffic flows from the source node to these destinations are then groomed (at the source node) and transmitted over a single light path to the different end nodes • 1xN passive optical splitters are placed at the optical terminations in order to extend the light path to N destinations NOBEL WP2: Valladolid Oct. 13-15, 2004

  32. 1, 2 and 3 are used Multicast approach (II): Example Unicast approach: a new light path is established for each requested connection Up to 4 connections can be allocated on each wavelength 4 5 Upcoming calls to be allocated 2 (1,6)  1 3 (1,7)  2 6 (1,5)  3 1 7 9 8 NOBEL WP2: Valladolid Oct. 13-15, 2004

  33. Only 1 is used Passive optical splitting Multicast approach (II): Example Multicast-like approach 4 5 2 Upcoming calls to be allocated 3 6 (1,6)  1 S (1,7)  1 1 (1,5) 1 7 9 8 NOBEL WP2: Valladolid Oct. 13-15, 2004

  34. Multicast approach (III) • Applying this strategy pursues two main objectives: • Optimisation of the optical channel capacity utilization by grooming at the source node different traffic flows towards different destination nodes into a single wavelength • Resources saving: when compared to a unicast scheme, the total number of transceivers required is lower as optical channels are re-used by allocating different client traffic flows on them • But,passive optical splitters are needed at each termination of the light paths NOBEL WP2: Valladolid Oct. 13-15, 2004

  35. Simulation case study (I) • The aim of the simulation study has been: • To find out the network conditions (ratio between the client connections required bandwidth and the wavelength capacity, sublambdas) for which the multicast approach performs better than the unicast one • The number of allocated client connections (calls) given a blocking probability target was used as metric for comparison NOBEL WP2: Valladolid Oct. 13-15, 2004

  36. Simulation case study (II) • NSFNET backbone network was used • 14 optical network elements (OXCs) connected by optical fibre links • The calls are routed over a WDM network using an adaptive routing algorithm based on the shortest path algorithm • A cost function taking into account not only the number of hops but also the actual traffic load was defined • A first-fit scheme is used to select the wavelength on each link • Blocking probability was set to 1% • N (number of destinations) = 3 NOBEL WP2: Valladolid Oct. 13-15, 2004

  37. Without WC With WC Simulation results (I) Number of wavelengths per link: 12 As the number of sublambdas increases (>12), the multicast approach performs better (it allows a higher number of allocated connections) NOBEL WP2: Valladolid Oct. 13-15, 2004

  38. Without WC With WC Simulation results (I) Number of wavelengths per link: 12 When the number of sublambdas is lower than 12, the unicast approach performs better since fewer resources at the optical level are used (no power splitting) NOBEL WP2: Valladolid Oct. 13-15, 2004

  39. Without WC With WC Simulation results (III) Wavelength per link: 12 Average wavelength capacity utilisation: number of the allocated client calls per wavelength was obtained Under the appropriate network conditions, the number of connections is doubled NOBEL WP2: Valladolid Oct. 13-15, 2004

  40. Simulation results (IV) • The cost savings provided by the multicast approach have been quantified • Specifically, the number of optical transmitters when the carried traffic is the same for both approaches has been calculated • Multicast approach allows savings which ranges from 60% to 85% • Even considering that the nodes will be more complex, there is an important reduction in the number of optical transmitters needed to provide the same efficiency NOBEL WP2: Valladolid Oct. 13-15, 2004

  41. Conclusions • The multicast approach is useful under certain network conditions • It depends on the number of wavelengths per fibre link and the ratio between the wavelengths capacity and the client connection bandwidth (sublambdas) • When the number of sublambdas is low unicast performs better • There is a ratio starting from which multicast outperforms unicast approach allowing the need of a fewer number of optical transmitters NOBEL WP2: Valladolid Oct. 13-15, 2004

  42. Future Work Study how to physically implement the multicast approach To find out the ratio between granularities and optimal N To analyze different algorithms to implement the multicast approach Simulate different traffic patterns Application of the multicast-like approach to build OVPN NOBEL WP2: Valladolid Oct. 13-15, 2004

More Related