Solving Routing and Spectrum Allocation Related...
Transcript of Solving Routing and Spectrum Allocation Related...
![Page 1: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/1.jpg)
ECOC 2013 Th.1.E.1
Solving Routing and Spectrum Allocation Related
Optimization Problems Luis Velasco, Alberto Castro, Marc Ruiz
Optical Communications
Group
![Page 2: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/2.jpg)
2 Th.1.E.1 Luis Velasco
Outline
I. RSA Basics
II. Solving Techniques
III. Advanced RSA
IV. Off-line Network Planning
V. In-operation Network Planning
![Page 3: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/3.jpg)
Part I RSA basics
Optical Communications
Group
Luis Velasco
![Page 4: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/4.jpg)
4 Th.1.E.1 Luis Velasco
Flexgrid
Flexgrid uses a finer spectrum granularity. The optical spectrum is divided into frequency slices (e.g. 6.25GHz).
It brings features that are not offered by the fixed grid networks, such as flexible bandwidth allocation. transporting optical connections with a capacity beyond 100Gb/s elasticity against time-varying traffic.
00:00 02:00 04:00 06:00 08:00 10:00 12:00 14:00 16:00 18:00 20:00 22:00 24:00 Time Period (hh:mm)
A B C D
12.5 GHz
25 GHz 37.5 GHz 75 GHz
37.5 GHz 50 GHz 50 GHz
A C D
![Page 5: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/5.jpg)
5 Th.1.E.1 Luis Velasco
Spectrum allocation
s1 s2 s|S| Frequency Slices(S)
Optical spectrum
Frequency
slice width
Central Frequency (cf)
Frequency
Guard band
Conn. c Conn. a Conn. b
Spectrum allocation entail dealing with two constraints: • spectrum continuity along the links of a given routing path: the same slices
must be used in all links of the path, • spectrum contiguity: the allocated slices must be contiguous in the
spectrum.
![Page 6: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/6.jpg)
6 Th.1.E.1 Luis Velasco
Slices and Channels
INPUT S, d OUTPUT C(d) 1: 2: 3: 4: 5:
Initialize: C(d) ← 0[|S|-nd+1 x |S|] for each i in [0, |S|-nd] do
for each s in [i, i+nd-1] do C(d)[s]=1
return C(d)
C(d) pre-computation
Set of channels C(d)
s1 s2 s3 s4 s5 s6 s7
s1 s2 s3 s4
s2 s3 s4 s5
s3 s4 s5 s6
s4 s5 s6 s7
s5 s6 s7 s8
Channels
c1∈C(a) c2∈C(b) c3∈C(c) Frequency
![Page 7: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/7.jpg)
7 Th.1.E.1 Luis Velasco
Basic RSA Problem Statement
From To BW
Demand Matrix
Flexgrid
Lightpaths
Given: 1. a set N of locations and a set of optical fibers E
connecting those locations; 2. the characteristics of the optical spectrum (i.e.,
spectrum width, frequency slice width) and the set of modulation formats;
3. a traffic matrix D with the amount of bitrate exchanged between each pair of locations in N;
Output: the Route and Spectrum Allocation for each demand in D.
Objective: one or more among:
Minimize the amount of bitrate blocked, Minimize the total amount of used slices, Minimize the total number of links used, etc.
![Page 8: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/8.jpg)
8 Th.1.E.1 Luis Velasco
Link-path Channel Assignment Formulation
* L. Velasco, et al., "Modeling the Routing and Spectrum Allocation Problem for Flexgrid Optical Networks," Springer Photonic Network Communications, 24, 177-186, 2012
xd Binary. Equal to 1 if demand d is rejected, 0 otherwise. ypc Binary. Equal to 1 if channel c is assigned to path p and 0 otherwise
P(d) Set of predefined candidate paths for demand d. C(d) Set of channels for demand d.
Pre-computed Parameters
Variables
δpe Equal to 1 if path p uses link e, 0 otherwise. γcs Equal to 1 if channel c includes slice s, 0
otherwise.
O(|P(d)|·|D|·|C|) variables O(|D|+|E|·|S|) constraints
Com
puta
tion
Tim
e (h
ours
)
Number of demands
0
2
4
6
8
10
100 150 200
k=10k=15k=20
TEL
Solved using CPLEX
![Page 9: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/9.jpg)
9 Th.1.E.1 Luis Velasco
Topology Design as a RSA Problem
Given: 1. a connected graph G(N,E), where N is the set of locations and E the
set of optical fibers; 2. the characteristics of the optical spectrum and modulation formats; 3. a traffic matrix D;
Output: 1. The route and spectrum allocation for each demand in D. 2. The links that need to be equipped;
Objective: Minimize number of links to be equipped to transport the given traffic matrix.
![Page 10: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/10.jpg)
10 Th.1.E.1 Luis Velasco
Node-link CA Formulation
* L. Velasco, et al., "Modeling the Routing and Spectrum Allocation Problem for Flexgrid Optical Networks," Springer Photonic Network Communications, 24, 177-186, 2012
wdec Binary. Equal to 1 if demand d uses channel c in link e, 0 otherwise ze Binary. Equal to 1 if link e is opened, 0 otherwise
Variables
O(|D|·|E|·|C|) variables O(|D|·|C|·|V|·|E|) constraints
![Page 11: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/11.jpg)
11 Th.1.E.1 Luis Velasco
Network Dimensioning as a RSA Problem
Given: 1. a connected graph G(N,E), where N is the set of locations and E the
set of optical fibers; 2. the characteristics of the optical spectrum and modulation formats; 3. a traffic matrix D; 4. the cost of every component, such as optical cross-connects
(OXC) and transponder (TP) types specifying its capacity and reach.
Output: 1. The route and spectrum allocation for each demand in D. 2. Network dimensioning including the type of OXC and TPs in each
location;
Objective: Minimize the total cost to transport the given traffic matrix.
![Page 12: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/12.jpg)
Part II Solving Techniques
Optical Communications
Group
Luis Velasco
![Page 13: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/13.jpg)
13 Th.1.E.1 Luis Velasco
Heuristics
RSA problems are NP-hard, so there is little hope of ever finding efficient exact solution procedures for them.
Un-tractability of MILP formulations appears when instances to be solved involve a large number of variables. Tens of nodes x Tens of links x Hundreds of slots x Hundreds of demands =
Millions of binary variables.
Heuristics, i.e., approximate solution techniques, can be used to tackle RSA-based (combinatorial) problems.
Ω Solution space f: Ω→R Objective function defined on the solution space goal: find ω* ∈ Ω, f(ω)≥f(ω*) ∀ ω ∈ Ω
![Page 14: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/14.jpg)
14 Th.1.E.1 Luis Velasco
Heuristics
Heuristic algorithms usually consists on two phases: Constructive Phase, where a solution is built. Local search, where the solution is improved.
During the Constructive phase, greedy algorithms may be used.
During the local search phase, exchanges among elements in the solution and not in the solution are done.
N(ω)
ω
ω'
![Page 15: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/15.jpg)
15 Th.1.E.1 Luis Velasco
Meta-heuristics
Meta-heuristics allow go beyond heuristics by: adding variability (randomize) allowing escaping from local optima, at risk of cycling
Some well-known meta-heuristics are: GRASP (Feo and Resende): a multi-start metaheuristic for
combinatorial problems. Evolutionary algorithms (genetics). BRKGA (M. Resende) Simulated Annealing, probabilistic metaheuristic often used when
the search space is discrete Tabu Search (Fred W. Glover): Enhances the performance of Local
search by using memory structures. Ant colony: probabilistic technique (Marco Dorigo) Path relinking: an intensification method.
![Page 16: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/16.jpg)
16 Th.1.E.1 Luis Velasco
Hybrid meta-heuristics Heuristic hybridization: Combine several of the previous techniques.
GRASP + PR -> Diversification + Intensification
Example: multi-greedy + PR Three different constructive algorithms to provide diversification. Path Relinking finds new solutions in the path connecting two solutions.
ES = incumb= Ø ite = 1
Greedy Algorithm - 1
Greedy Algorithm - 2
Greedy Algorithm - 3
Local search
PR (Solution <-> ES)
Update incumb and
ES
Select at random {1,2,3}
ite++ stop?
yes no
1
2 3
* M. Ruiz, et al., "Ultra-fast meta-heuristic for the spectrum re-allocation problem in flexgrid optical networks", 25th European Conference on Operational Research (EURO XXV), 2012
Solution
end
![Page 17: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/17.jpg)
17 Th.1.E.1 Luis Velasco
Hybrid meta-heuristics: Solving Time
Time (seconds)
Opt
imal
ity g
ap
G3 provides feasible solutions in few
miliseconds (20-50 msec)
Multi-start with PR provides the best results:
(G1 + G3) (G2 + G3)
(G1 + G2 + G3)
G1 and G2 provide better solutions but at the expense of higher computation time
100 msec 1 sec
![Page 18: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/18.jpg)
18 Th.1.E.1 Luis Velasco
Large Scale Optimization (LSO)
The objective of LSO methods is to extend the exact methodology (Branch & Bound) for MILP formulations.
Among different methods, decomposition methods have been successfully used for solving communications network design problems Lagrangean relaxation aims at improving lower bounds on the objective
function, a decisive factor for the effectiveness of Branch & Bound. Column generation consists in finding a reduced set of variables
(columns) to solve the linear relaxation of the link-path MILP formulations, providing high quality integer solutions in an efficient way.
Benders decomposition is an iterative procedure based on projecting out a subset of variables from the original problem with the whole set of variables and creating new constraints (cuts) from the projected ones.
![Page 19: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/19.jpg)
19 Th.1.E.1 Luis Velasco
Lagrangian Relaxation
Some (difficult) constraints are moved to the objective function and penalized by Lagrange multipliers (λ)
∑∈
⋅Dd
dd bxmin
Ddxy ddPp dCc
pc ∈∀=+∑ ∑∈ ∈
1)( )(
SsEeyDd dPp dCc
pcpecs ∈∈∀≤⋅⋅∑ ∑ ∑∈ ∈ ∈
,1)( )(
δγ
∑∑ ∑ ∑ ∑∑∈ ∈ ∈ ∈ ∈∈
−⋅⋅⋅+⋅
Ee Ss Dd dPp dCcpcpecses
Dddd ybx
)( )(1min δγλ
Ddxy ddPp dCc
pc ∈∀=+∑ ∑∈ ∈
1)( )(
s.t.
s.t.
It allows obtaining lower bounds for the original problem for any positive Lagrange multipliers.
Those Lagrange multipliers that provide the optimal solution of the original problem can be found by solving the Lagrangian dual problem. Iterative methods such as sub-gradient optimization methods are
commonly used for solving the Lagrangian dual problem.
![Page 20: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/20.jpg)
20 Th.1.E.1 Luis Velasco
Column Generation Algorithm
Initial set of lightpaths P
Solve RSA (Master
Problem, LP)
Solve Pricing Problem
Dual variables
New lightpaths
added
Solve RSA (ILP)
No more lightpaths found
Main Algorithm Lightpath set initialization
* M. Ruiz, et al., “Column Generation Algorithm for RSA Problems in Flexgrid Optical Networks,” Springer Photonic Network Communications, 2013.
For each demand d: 1) Compute the shortest (hops) path in the network 2) Generate a lightpath with the shortest path and the first channel
for demand d. Pricing Problem 1) For each demand d and each channel in Cd, compute the
shortest path in the network with link metrics:
2) Return those lightpaths (route + channel) with the highest positive reduced costs (if any)
Exhaustive search of channels optimized by means of Floyd-Warshall algorithm
dcSs
ese Cccf ∈= ∑∈ )(
)( π
0
5000
10000
15000
20000
25000
30000
35000
40000
32 40 48 56
ILP (K=5)
Column GenerationNum
ber o
f var
iabl
es
1.E-3
1.E-2
1.E-1
1.E+0
1.E+1
1.E+2
1.E+3
32 40 48 56
ILP (K=5)
Column Generation
# demands
0.14 0.4
solv
ing
time
(min
)
141.7
# demands
![Page 21: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/21.jpg)
21 Th.1.E.1 Luis Velasco
The Dijkstra’s Shortest Path algorithm
Each node i is labeled with the aggregated metric d(i) from the source node and with its predecessor pred(i).
the route source-i (subset of links E(source, i) ⊆ E) can be computed visiting the predecessor node starting from i, until source node is reached.
Procedure Dijkstra (N, E, source) begin S : = {source}, T = N – {source} d(source) = 0 and pred(source) = 0 d(j) = ∞ for each j ≠ source update(source) while S ≠ N do
let i ∈ T be a node for which d(i) = min {d(j) : j ∈ T} S = S ∪ {i}, T = T – {i} update(i)
end Procedure Update (i) for each (i,j) ∈ E(i) do
if d(j) > d(i) + cij then d(j) = d(i) + cij pred(j) = i
![Page 22: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/22.jpg)
22 Th.1.E.1 Luis Velasco
Dijkstra-based RSA (1/2)
The label also contains ηs(i), the aggregated state of frequency slice s
ηs(i) = ∏eϵE(source,i) ηes ∀ sϵS
The downstream node j of node i updates the label only if at least one channel is available, computing:
∈∀=⋅∈∃
==otherwise0
)(1)(:)(1)),((
cSsidCcjie ess ηη
π
Procedure Update (i) for each (i,j) ∈ E(i) do
if π(i,j)==1 and d(j) > d(i) + cij then
d(j) = d(i) + cij pred(j) = i
i
η(i) ηe3
ηe2
ηe1
src
![Page 23: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/23.jpg)
23 Th.1.E.1 Luis Velasco
Dijkstra-based RSA (2/2)
Spectrum allocation can be done implementing any heuristic, such: First Fit Random Least fragmented spectrum etc.
Modulation formats, reachability, etc, can be easily included.
![Page 24: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/24.jpg)
Part III Advanced RSA
Optical Communications
Group
Luis Velasco
![Page 25: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/25.jpg)
25 Th.1.E.1 Luis Velasco
BV-OXC
MF-TPs
Flexgrid-based Core
Network
Client flows
Source of traffic variations
Reducing aggregation level might increase traffic variations
Intra-datacenter network
Flexgrid-based Network
Scheduler
Cloud Middleware
QoS estimation
Optical Connection Datacenter
A
IT resources
Scheduler
Cloud Middleware
QoS estimation
Intra-datacenter network
Datacenter B
IT resources Active Stateful
PCE
ABNO Controller
Policy Agent
TED LSP-DB
Prov. Mngr OAM Handler
Datacenter interconnection to transfer bulk data
* L. Velasco, et al., "Cross-Stratum Orchestration and Flexgrid Optical Networks for Datacenter Federations," accepted in IEEE Network Magazine, 2013.
![Page 26: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/26.jpg)
26 Th.1.E.1 Luis Velasco
Shared Spectrum
t
t+1
Allocated spectrum
t
t+1
Allocated spectrum Used spectrum
t
t+1
Elastic CF Allocated spectrum
Elastic Spectrum Allocation Policies
Fixed: both the assigned CF and spectrum width do not change in time.
* M. Klinkowski, et al., "Elastic Spectrum Allocation for Time-Varying Traffic in Flexgrid Optical Networks," IEEE Journal on Selected Areas in Communications (JSAC), vol. 31, pp. 26-38, 2013
Semi-Elastic: the assigned CF is fixed but the allocated spectrum may vary. • At each time interval, the allocated
spectrum corresponds to the utilized spectrum.
• Spectrum increments/decrements are achieved by allocating/releasing frequency slices.
• The frequency slices can be shared between neighboring demands, but used by, at most, one demand in a time interval.
Elastic: both the assigned CF and the spectrum width can be subject to change in each time interval.
![Page 27: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/27.jpg)
27 Th.1.E.1 Luis Velasco
Multi-period RSA
Find, for each demand in D: a route and a SA for each time period in T, constraining SA changes to a certain policy.
Complexity depends on SA policy SA between consecutive time periods is constrained by the chosen policy
RSA
MP-RSA
![Page 28: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/28.jpg)
28 Th.1.E.1 Luis Velasco
Dynamic lightpath adaptation
Given: a core network topology represented by a graph G(N, E); a set S of available slices of a given spectral width for every link in E; a set L of lightpaths already established on the network; each lightpath l is
defined by the tuple {Rl, fl, sl}, where the ordered set Rl ⊆ E represents its physical route, fl its central frequency and sl the amount of frequency slices.
a lightpath p ∈ L for which spectrum adaptation request arrives and the required number of frequency slices, (sp)req.
Output: the new values for the spectrum allocation of the given lightpath p: {Rp, fp,
(sp)’} and {Rp, (fp)’, (sp)’}, respectively, if the Semi-Elastic and Elastic policy is used.
Objective: maximize the amount of bit-rate served.
![Page 29: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/29.jpg)
29 Th.1.E.1 Luis Velasco
INPUT: G(N,E), S, L, p, (sp)req OUTPUT: (sp)’
1: 2: 3: 4: 5: 6: 7: 8: 9:
if (sp)req ≤ sp then (sp)’ ← (sp)req else
L+ ← Ø, L- ← Ø for each e ∈ Rp do
L- ← L- ∪ {l ∈ L: e ∈ Rl , adjacents(l, p) , fl<fp} L+ ← L+ ∪ {l ∈ L: e ∈ Rl , adjacents(l, p) , fl>fp}
smax ← 2*min{min{fp – fl – sl, l ∈ L-}, min{fl – fp – sl, l ∈ L+}} (sp)’ ← min{smax, (sp)req}
return (sp)’
On-line Algorithms for Elastic SA
INPUT: G(N,E), S, L, p, (sp)req OUTPUT: (fp)’, (sp)’
1: 2: 3: 4: 5: 6: 7: 8: 9:
10:
if (sp)req ≤ sp then (sp)’ ← (sp)req; (fp)’ ← fp else
L+ ← Ø, L- ← Ø for each e ∈ Rp do
L- ← L- ∪ {l ∈ L: e ∈ Rl , adjacents(l, p) , fl<fp} L+ ← L+ ∪ {l ∈ L: e ∈ Rl , adjacents(l, p) , fl>fp}
smax ← min{fp – fl – sl, l ∈ L-} + min{fl – fp – sl, l ∈ L+} (sp)’ ← min{smax, (sp)req} (fp)’ ← findSA_MinCFShifting (p, (sp)’, L+, L-)
return {(fp)’, (sp)’}
Algorithm for the semi-elastic policy
Algorithm for the elastic policy
* A. Asensio, et al., "Impact of Aggregation Level on the Performance of Dynamic Lightpath Adaptation under Time-Varying Traffic," in Proc. IEEE International Conference on Optical Network Design and Modeling (ONDM), 2013.
![Page 30: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/30.jpg)
30 Th.1.E.1 Luis Velasco
Example of Re-optimisation (Defragmentation)
A B C
Re-allocation
A B C
Connection set-up
A B C
Initial Spectrum
25 GHz 50 GHz 37.5 GHz
37.5 GHz
![Page 31: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/31.jpg)
31 Th.1.E.1 Luis Velasco
Spectrum Reallocation (SPRESSO)
* A. Castro, et al., "Dynamic Routing and Spectrum (Re)Allocation in Future Flexgrid Optical Networks," Elsevier Computers Networks, vol. 56, pp. 2869-2883, 2012.
xpc binary, 1 if channel c is assigned to path p, 0 otherwise. yp binary, 1 if path p is reallocated, 0 otherwise. r integer with the number of already established paths to reallocate.
Variables Pm subset of P with the candidate paths, including newP. Pm = P(newP) ⋃ {newP}.
Parameters
O(|Pm|·|C|) variables O(|Pm|·|C|+|E|·|S|) constraints
![Page 32: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/32.jpg)
32 Th.1.E.1 Luis Velasco
Run RSA algorithm
Connection request d
arrives
Solution found?
Establish path
Block connection
request
Run Re-optimization algorithm
Yes No
Perform Re-optimization
Solution found?
No
Yes
End
Start
Provisioning-triggered re-optimization
![Page 33: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/33.jpg)
33 Th.1.E.1 Luis Velasco
Gains of Using SPRESSO
Offered load (Erlangs)
Pbbw
(%)
TEL BT DT
0
1
2
3
4
5
20 40 60 80 100 120 140 160 180 20 40 60 80 100 120 140 20 30 40 50 60 70 80
50GHz 50GHz (w/ SPRESSO)25GHz 25GHz (w/ SPRESSO)12.5GHz 12.5GHz (w/ SPRESSO)6.25GHz 6.25GHz (w/ SPRESSO)
Network 25 GHz 12.5 GHz 6.25 GHz
TEL 30,4% 28,4% 30,3% BT 22,9% 23,0% 34,1% DT 22,2% 24,8% 33,0%
TEL
BT
DT
* A. Castro, et al., "Dynamic Routing and Spectrum (Re)Allocation in Future Flexgrid Optical Networks," Elsevier Computers Networks, vol. 56, pp. 2869-2883, 2012.
![Page 34: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/34.jpg)
34 Th.1.E.1 Luis Velasco
Bulk path computation
Path computation for a set of connection requests. the bulk of path requests is computed attaining the optimal solution for the
whole set. Increases optimality but increases also provisioning time.
Bulk path computation can be used for restoration.
* R. Martínez, et al., "Experimental Validation of Dynamic Restoration in GMPLS-controlled Multi-layer Networks using PCE-based Global Concurrent Optimization," in Proc. IEEE/OSA Optical Fiber Communication Conference (OFC), 2013.
PCE GCO
Queue
Path Computation requests for restoration
Reduces resource contention. Increases resource utilization, specially in
MLN. Stringent computation times require
heuristic algorithms.
![Page 35: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/35.jpg)
35 Th.1.E.1 Luis Velasco
Bitrate Squeezed and Multi-path Recovery
s
1 2
t
5 6 7
3 4
restoration paths
s
1 2
t
5 6 7
3 4
working path
backup paths (reserved)
6 slices
2 slices
Path Recovery Description
Protection
Protection routes are known in advance:
• Dedicated Protection • Shared Protection
Restoration Restoration routes are found adaptively based on the failure and the state of the network at the time of failure.
2 slices
2 slices
2 slices
* A. Castro et al., "Single-path Provisioning with Multi-path Recovery in Flexgrid Optical Networks," International Workshop on Reliable Networks Design and Modeling (RNDM), 2012
![Page 36: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/36.jpg)
Part IV Off-line Network Planning
Optical Communications
Group
Luis Velasco
![Page 37: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/37.jpg)
37 Th.1.E.1 Luis Velasco
Network Life Cycle
Network planning is performed periodically:
Planning &
Forecast
Architect & Design
Implement design
Network Operation
Monitor and
Measure
New Services Population grow Capacity is installed to guarantee
that the network can support the forecast traffic.
Long planning cycles are used to upgrade the network and prepare it for the next planning period.
Results from network capacity planning are manually deployed in the network.
![Page 38: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/38.jpg)
38 Th.1.E.1 Luis Velasco
Extending the core towards the metro
Migrating Towards Flexgrid Technology
core metro border nodes
Enlarging the core Sub-network Upgrading
1 2 3
* M. Ruiz, et al, "Planning Fixed to Flexgrid Gradual Migration: Drivers and Open Issues," submitted to IEEE Communications Magazine, 2014.
![Page 39: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/39.jpg)
39 Th.1.E.1 Luis Velasco
Migrating connections (1/4)
A B
A B Fixed grid
![Page 40: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/40.jpg)
40 Th.1.E.1 Luis Velasco
Migrating connections (2/4)
A B
A B Flexgrid
![Page 41: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/41.jpg)
41 Th.1.E.1 Luis Velasco
Migrating connections (3/4)
A B C
A B C Mixed grid
![Page 42: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/42.jpg)
42 Th.1.E.1 Luis Velasco
Migrating connections (4/4)
A B C D
A B
C D
Mixed grid
![Page 43: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/43.jpg)
43 Th.1.E.1 Luis Velasco
Migration flow chart
Network Reconfiguration
Planning Tool
Planning Department
Inventory
Engineering Department
NMS
yes
no Purchasing,
Installing, Reconfiguring,
Testing
Requirements met?
Planning requests Reconfiguration
requests
Network Upgrading Creating/ extending
flexgrid islands
Enlarging to un-deployed
areas
Extending to border metro
areas
![Page 44: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/44.jpg)
Part V In-operation Network Planning
Optical Communications
Group
Luis Velasco
![Page 45: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/45.jpg)
45 Th.1.E.1 Luis Velasco
Static Network Operation
Metro (Vendor A)
Metro (Vendor B)
IP/MPLS (Vendor A)
IP/MPLS (Vendor B)
OXC (Vendor A)
OXC (Vendor B)
Internet Voice CDN Cloud Business
Umbrella Provisioning System
Metro OSS
IP/MPLS Core OSS
Optical Transport OSS
NMS (Vendor A)
NMS (Vendor B)
NMS (Vendor A)
NMS (Vendor B)
NMS (Vendor A)
NMS (Vendor B)
Service Management
Systems
Network Operations
Support System
Transport Network
Nodes
![Page 46: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/46.jpg)
46 Th.1.E.1 Luis Velasco
IETF architectures supporting in-operation Planning (1/2)
Architecture Strengths Weaknesses
Stateless PCE
• Path computation can be off-loaded onto a dedicated entity capable of complex computations with tailored algorithms and functions.
• Has a standard and mature interface and protocol.
• Supports simple optimisation, such as bulk path computation.
• Is unaware of existing LSPs and has no view of the current network resource utilisation and key choke points.
• Cannot configure by itself any LSP in the network.
• Delays need to be introduced to sequence LSP set-up.
Stateful PCE
• Maintains a database of LSPs that are active in the network, i.e. so that new requests can be more efficiently placed optimising network resources.
• Supports optimization involving already established LSPs.
• More complex than a stateless PCE, requires additional database and synchronization.
• No existing LSPs can be modified, e.g. for network re-configuration purposes.
![Page 47: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/47.jpg)
47 Th.1.E.1 Luis Velasco
IETF architectures supporting in-operation Planning (2/2)
Architecture Strengths Weaknesses
Active Stateful
PCE
• Capable of responding to changes in network resource availability and predicted demands and reroute existing LSPs for increased network resource efficiency.
• Supports complex reconfiguration and re-optimization, even in multilayer networks.
• No new LSPs can be created, e.g. for VNT re-optimisation purposes.
• Requires protocol extensions to modify and/or instantiate (if the capability is available) LSPs.
ABNO
• Provides a network control system for coordinating OSS and NMS requests to compute paths, enforce policies, and manage network resources for the benefit of the applications that use the network.
• New LSPs can be created for in-operation planning. VNTM in charge of VNT re-configuration.
• Supports deployment of solutions in multi-technology scenarios (NetConf, OpenFlow, control plane, etc.)
• Requires implementation of a number of key components in addition to the PCE function.
• Some interfaces still need to be defined and standardized.
![Page 48: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/48.jpg)
48 Th.1.E.1 Luis Velasco
Migration towards in-operation network planning
Metro (Vendor A)
Metro (Vendor B)
IP/MPLS (Vendor A)
OXC (Vendor A)
IP/MPLS (Vendor B)
OXC (Vendor B)
Internet Voice CDN Cloud Business
Application-based Network Operations (ABNO)
In-Operation Planning Tool
PCE
ABNO Controller
TED
LSP-DB
Prov. Mngr OAM Handler
VNTM
ABNO
North Bound Interface
South Bound Interface
Metro OSS
IP/MPLS Core OSS
Optical Transport OSS
GMPLS Control Plane
![Page 49: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/49.jpg)
49 Th.1.E.1 Luis Velasco
Extended Networks Life Cycle
Planning & Forecast
Architect & Design
Implement design
Network Operation
Monitor and
Measure
New Services Population grow
Reconfigure / Re-optimise
![Page 50: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/50.jpg)
50 Th.1.E.1 Luis Velasco
Re-optimisation Process
Active Stateful
PCE
ABNO Controller TED
LSP-DB
Prov. Mngr OAM Handler
Back-end PCE
TED LSP-DB
Active Solver PCEP
Server
PCEP
BGP-LS
1
11
4
5 6 7
8
9
10
Flexi-grid Core Network
OXC
Provisioning 3
12
NMS / OSS
ABNO
Router A Router B
In-operation Planning
Service Layer Inventory Databases
GMPLS Control Plane
2
* L. Velasco, et al., "In-operation Network Planning ," submitted to IEEE Communications Magazine, 2014.
![Page 51: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/51.jpg)
51 Th.1.E.1 Luis Velasco
In-Operation Planning Tool: PLATON
CPU GPU
* L. Gifre, et al. “Architecture of a Specialized Back-end High Performance Computing-based PCE for Flexgrid Networks," in Proc. IEEE International Conference on Transparent Optical Networks (ICTON), 2013.
![Page 52: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/52.jpg)
52 Th.1.E.1 Luis Velasco
Conclusions
Basic RSA algorithms have been reviewed
Elastic SA: Semi-elastic and Elastic policies
Off-line planning: and example for gradual migration process from fixed to Flexgrid has been presented.
Re-optimization can be performed to increase network performance.
A control and management architecture to support in-operation planning. The architecture allows dynamic network operation and to reconfigure and
re-optimise the network near real-time in response to changes. Networks life cycle is extended achieving better resource utilization. Process automation reduces manual interventions and, consequently, OPEX.
PLATON: an In-Operation Planning Tool has been presented.
![Page 53: Solving Routing and Spectrum Allocation Related ...personals.ac.upc.edu/lvelasco/docs/research/2013-ECOCTutorial.pdf · I. RSA Basics II. Solving Techniques III. Advanced RSA IV.](https://reader034.fdocuments.us/reader034/viewer/2022052005/601975c9be52bf479029dc0d/html5/thumbnails/53.jpg)
ECOC 2013 Th.1.E.1
Solving Routing and Spectrum Allocation Related Optimization Problems
Luis Velasco, Alberto Castro, Marc Ruiz [email protected]
Optical Communications
Group
Thank you for your attention!