Reconfiguration of Traffic Reconfiguration of Traffic Grooming - - PowerPoint PPT Presentation

Reconfiguration of Traffic Reconfiguration of Traffic Grooming Optical Networks Grooming Optical Networks

Ruhiyyih Mahalati and Rudra Dutta

Computer Science, North Carolina State University

This research was supported in part by NSF grant # ANI-0322107

Rudra Dutta, NCSU, BroadNets '04 presentation 2

l Context l Problem Definition l Integrated Approach Formulation l Reconfiguration Heuristic

– Over-Provisioning Methods – Hard & Soft Decision Criterion – Flowchart

l Numerical Results l Conclusion

Outline Outline

Rudra Dutta, NCSU, BroadNets '04 presentation 3

Virtual Topology, Traffic Grooming Virtual Topology, Traffic Grooming

Rudra Dutta, NCSU, BroadNets '04 presentation 4

• Certain

wavelengths pass through

• ptically
• Others

terminated at Digital Cross Connect (DXC) for OEO

Optical Cross Connect Optical Cross Connect

Rudra Dutta, NCSU, BroadNets '04 presentation 5

l Traffic Grooming: Combining lower speed traffic flows onto wavelengths to minimize network cost l Traffic Grooming problem conceptually comprises of

1. Virtual Topology SP 2. Routing & Wavelength Assignment SP 3. Traffic Routing SP

Traffic Grooming Traffic Grooming

Physical Topology Gp Virtual Topology Gv Traffic T Grooming G Routing R l assign L

Rudra Dutta, NCSU, BroadNets '04 presentation 6

l Reconfiguration: possibility of adaptively creating virtual topologies, based on network need

– Independence between the virtual and the physical topology

l Goal: Improve performance metric l Tradeoff between the performance metric value and the number of changes l Computationally intractable l Many practical heuristics exist

Reconfiguration Reconfiguration

Rudra Dutta, NCSU, BroadNets '04 presentation 7

Reconfiguring Groomed Networks Reconfiguring Groomed Networks

l Are existing methods sufficient to reconfigure with subwavelength traffic?

– If not, what are the needs?

l Observation: full wavelength reconfiguration cannot modify grooming of traffic onto virtual topology

– How to translate change of subwavelength traffic to change

• f lightpaths?

l Observation: reconfiguration cost is defined from considerations different from grooming

Physical Topology Gp Virtual Topology Gv Traffic T Grooming G Routing R l assign L

Rudra Dutta, NCSU, BroadNets '04 presentation 8

l Integrated Approach - reconfiguration of a topology as well as traffic assignment in a groomed network, with the objective to balance grooming gain and reconfiguration cost l Assumptions:

– Each node is equipped with an OXC and DXC – Physical links and lightpaths are directed – No wavelength converters ÆNo more than a single lightpath between two nodes ÆDisallowing bifurcated routing of traffic

Problem Definition Problem Definition

Rudra Dutta, NCSU, BroadNets '04 presentation 9

l Grooming cost is normally represented as total number of LTEs or total electronic switching l Reconfiguration cost is normally represented as the number of network equipments that require reconfiguration l Our Integrated Cost Calculation:

– Grooming Cost: total amount of electronic switching - total traffic weighted delay – Reconfiguration Cost: the number of OXCs and DXCs that need reconfiguration - total delay experienced by the traffic at these nodes – Both measure delay suffered by traffic

The Need for a Cost Function The Need for a Cost Function

Rudra Dutta, NCSU, BroadNets '04 presentation 10

• Matrix representation of each node’s switching state

Reconfiguration Cost Function Reconfiguration Cost Function

Rudra Dutta, NCSU, BroadNets '04 presentation 11

l Lightpath establishment - OXC, DXC l Different optical switching - only OXC l Lightpath termination and origination at a node - single change to both OXC and DXC.

Matrix Distance as Cost Function Matrix Distance as Cost Function

Rudra Dutta, NCSU, BroadNets '04 presentation 12

l Global Reconfiguration Cost Calculation Methods

– RC-I = Total no. of OXCs, Total no. of DXCs – RC-II = Total no. of OXC wavelength changes, Total no. of DXCs – RC-III = Total no. of OXC changes, Total no. of DXCs – RC-IV = Total no. of OXC changes, Total no. of DXC changes : linear

l Integrated Approach as an ILP

– Objective: Maximize (Grooming gain) g - (RC-IV) - d – g : relative weightage parameter: related to average delay between reconfigurations

– d : to prevent chattering

ILP Formulation ILP Formulation

Rudra Dutta, NCSU, BroadNets '04 presentation 13

l Integrated Approach Solution as an ILP - optimal but computationally expensive

– Note: Optimal in the next state

l The heuristic approach must

– Avoid resorting to the full ILP whenever possible – Ward off failure of the network - remain feasible – Avoid adopting very suboptimal grooming solutions

l Problem is intractable - tractable heuristic unlikely to attain globally optimal solutions l Heuristic is proactive: over-provisioning

Proposed Heuristic Algorithm Proposed Heuristic Algorithm

Rudra Dutta, NCSU, BroadNets '04 presentation 14

l Model: traffic components are relatively static, but may change somewhat over time (LCAS)

– For revenue, increases are desirable to serve, decreases are desirable to leverage – For resilience, need to react fast to opportunities

l Over-provisioning at traffic demand level: use extra capacity,

• therwise unutilized

l OXCs and DXCs configured to carry over-provisioned traffic l Family of traffic matrices supported

– All new traffic matrices that are subset of the initial traffic matrix

l Lightpath slack limits over-provisioning

– Equal allocation – Prorated allocation – Inverse allocation

Over-provisioning Over-provisioning

Rudra Dutta, NCSU, BroadNets '04 presentation 15

l Different Methods of Over-Provisioning

– Equal over-provisioning method

l Pick minimum over-provisioned over all traffic elements

– Selective over-provisioning method

l Pick minimum over-provisioned for each individual traffic element

– Iterative over-provisioning method

l Iteratively over-provision some traffic elements with any extra capacity, if available l Several variants possible l Similar performance for the variants

Over-provisioning Approaches Over-provisioning Approaches

Rudra Dutta, NCSU, BroadNets '04 presentation 16

C = 15

3,7,2 Over-provision 1,1,1

Over-provisioning Example Over-provisioning Example

Rudra Dutta, NCSU, BroadNets '04 presentation 17

Over-provisioning Strategies Over-provisioning Strategies

l Equal

– Every t(sd) gets the same (therefore min) - simplistic

l Selective

– Every t(sd) gets the max they can get

l Iterative

– One t(sd) is assigned its max, then slacks recalculated – Different flavors depending on the choice

l Iterative-Min l Iterative-Max l Iterative-Ratio l Iterative-Max-lightpath l Iterative-Min-Max

Rudra Dutta, NCSU, BroadNets '04 presentation 18

Over-provisioning comparison Over-provisioning comparison

Rudra Dutta, NCSU, BroadNets '04 presentation 19

l Traffic change - grooming cost may increase - reconfiguration needed

– But very frequent reconfigurations undesirable

l Critical region: sub-wavelength elements carrying traffic close to over-provisioned traffic (threshold)

– reconfiguration triggered

l LPlimit: ratio of lightpaths carrying sub-wavelength elements in critical region

– LPlimit decides hard or soft decision criterion

l Hard Decision: global reconfiguration

– Integrated ILP

l Soft Decision: local reconfiguration

– only DXC reconfiguration

Heuristic Description Heuristic Description

Rudra Dutta, NCSU, BroadNets '04 presentation 20

Heuristic Flowchart Heuristic Flowchart

Rudra Dutta, NCSU, BroadNets '04 presentation 21

l Traffic Evolution - Rising, Falling, Rising & Falling l Parameters: g = 2, 7, 15, 200, LPlimit = 30%, 70% l “Grooming-only”, Integrated approach, Heuristic

– Reconfiguration Cost

– Grooming Cost – Integrated Objective – Cumulation of the Integrated Objective

• Given: a physical

topology, initial traffic matrix, a series of changing traffic matrices

• 4 Physical Topologies

Numerical Results Numerical Results

Rudra Dutta, NCSU, BroadNets '04 presentation 22

Reconfiguration Cost Reconfiguration Cost

Rudra Dutta, NCSU, BroadNets '04 presentation 23

Grooming Cost Grooming Cost

Rudra Dutta, NCSU, BroadNets '04 presentation 24

Integrated Objective Integrated Objective

Rudra Dutta, NCSU, BroadNets '04 presentation 25

Cumulation Cumulation of Integrated Objective

• f Integrated Objective

Rudra Dutta, NCSU, BroadNets '04 presentation 26

More Numerical Results More Numerical Results

Rudra Dutta, NCSU, BroadNets '04 presentation 27

More Numerical Results More Numerical Results

Rudra Dutta, NCSU, BroadNets '04 presentation 28

l Proposed a new problem - joint grooming and reconfiguration l Defined basis for comparison - provided integrated cost function l Problem formulation as an ILP l Heuristic – robust to variation in physical topology

– Integrated approach - maximum integrated objective – Cumulation of integrated objective - heuristic follows integrated approach while “Grooming-only” approach deviates – “Grooming-only” approach - not suitable for reconfiguration of groomed traffic

l Heuristic considerably reduces ILP calculation l LPlimit reduced - Heuristic performance improves l Very high g - Integrated approach gives optimal grooming cost, still incurring less reconfiguration cost l Verified through numerical experiments

Summary and Conclusion Summary and Conclusion