Traffic Grooming : The intelligent allocation of traffic demands onto an available set of wavelengths in a way that reduces the overall cost of the network. The Traffic Grooming Probl[r]
(1)Survivable Network Design Survivable Network Design
David Tipper
Associate Professor
Associate Professor
Department of Information Science and Telecommunications
University of Pittsburgh
Telcom
Telcom2110 Slides 152110 Slides 15
Survivable Network Design
• Spare Capacity Allocation (SCA) Problem:
– given working paths and network (or virtual network) topology – provision spare capacity and find backup routes for fault tolerance
– Goal: minimumspare capacity or cost
• Survivable Mesh Networks
– Consider preplanned protectionin mesh networks • STM - DCS, ATM - VP, WDM, MPLS, etc
(2)Classification of Survivability Techniques
• Path-based (Global) versus Link-based (Local) • Failure Dependent vs Failure Independent • Protection versus Restoration
• Dedicated-Backup versus Shared- Backup Capacity • Ring versus Mesh topology
• Dual and multi-homing • P cycle
• Etc.
Failure Dependent vs Failure Independent
• Failure Dependent – the backup path depends on which device fails – need a set of paths one for each failure case • Failure Independent – backup path link and node disjoint
with working path - one backup path per working path • Example:
13
12
10
9
2
7
3
5
Working path
Failure Dependent backup path for link 1-2 failure
(3)SCA Problem
• SCA for Failure Independent Shared Backup Path Restoration
• Notation
r = 1,2,…, D set of demands (source-destination pairs)
p = 1,2,…, Pr set of possible paths for demand pair r
l = 1,2,…, L set of network links • Input parameters (constants)
α r offered traffic load of demand pair r
cl unit cost of capacity on link l
δl r,p = 1 if lbelongs to path prealizing demand r
= 0,otherwise
f set of link failure scenarios • variables
x r,p flow of demand ron path p
sl spare capacity on link l
SCA Path-flow model
Find sl and x r,p , which
∑ ∈ ⋅ L l l l s c minimize D r x r P p p
r = ∀ ∈
∑ ∈ , 1 , L f f f L l s x l D
r p P
p r p r l r f r ∈ ∀ − ∈ ∀ ≤ ⋅ ∑ ∑ ∈ ∈ , }, { , , , δ α s.t.
Total spare capacity
Single backup path for each flow
(4)Matrix Based Formulation of SCA
• Matrix Based formulation of Optimization model for FID shared backup path restoration*
• Consider path incident matrices Pand Qfor working and backup paths where each matrix has
number of rows = number of flows in the network number of columns = number of links in the network – row iin the matrix Pcorresponds to the set of links used by flowi – where pij= 1if flow iuses link jit is 0otherwise
– similary row iin the matrix Qcorresponds to the set of backup path links used by flowi where qij= 1if flow iuses link jit is 0otherwise
• Relate to spare provision matrix G, and spare capacity reservation s
– G= QT P, element G
ijgives required spare capacity on link iwhen link
jfails
– s= max(G), or s≥G , spare capacity reservations are the maximum spare capacity for any single link failure
• * Y.Liu, D.Tipper, and P Siripongwutikorn, “Approximating Optimal Spare Capacity Allocation by Successive Survivable Routing,'' ACM/IEEE Transactions on Networking, Vol 13., No 1, pp 198-211, Feb., 2005
Example
Link i 1 2 3 4 5 6 7
Backup path link incident matrix
1 2 1 1 0 1 1 1 0
2 2 1 0 1 0 1 0
3 1 0 1 0 0 0
4 1 1 0 1 0 1 0
5 1 0 1 0 0 0
6 0 1 0 0 0 1
7 2 0 0 1 0
11
Flows 3 45 10
src dst 0 0 0 a b 1 0 0 a c 1 0 0 3 a d 1 0 0 4 a e Working path link 0 0 0 b c incident matrix 0 1 0 b d 0 0 b e 0 0 0 c d
From G,
s=maxG
From working and backup paths, G= QT P
P
QT
G s
An example: when link fails,
3
4
5
a
c b
(5)Matrix Based SCA for Link Failures
min S = eTs
Q,s
s.t. s≥ G
G = QTM P
P + Q ≤ 1
Q BT= D (mod 2)
Qis a binary matrix Decision variable: Q, s
Given: M – traffic demand matrix
P – working path link incidence matrix
Band D– node-link & flow-node incidence matrices
Total spare capacity
Link-disjointed backup paths Flow conservation of backup
Integer programming Calculation of spare provision matrix Enough spare capacity on each link
Another way to find G
G = ΣrGr, where Gr= qrTp
r, pr and qrare
vectors for working and backup paths of flow r
G2 G G1
GR GR-1
…
P
Q
(6)The Traffic Grooming Problem
• Number of wavelengths per fiber = -100+ • Per wavelength capacity = 2.5 Gbps to10 Gbps • Bandwidth requirements of most applications << 2.5
Gbps
∴Group several sessions on the same wavelength channel in order to better utilize the available bandwidth
Traffic Grooming: The intelligent allocation of traffic demands onto an available set of wavelengths in a way that reduces the overall cost of the network.
The Traffic Grooming Problem: CapEx
• Dominant cost factor: Electronic layer multiplexing; number of electronic layer Line Terminating Equipment (LTs):
– SONET/SDH ADMs – IP/MPLS router ports
• Solution:Assign the traffic such that minimum number of LTs is used
3-4 times as expensive as OXC
(7)Traffic Grooming Problems
• Network design problem: dimensioning and network provisioning
– Reduce capital and operational expenditure – Maximize revenue
NP-Complete Problem
• Solution types:
– Exact solutions (based on ILP or MILP) – Heuristic and approximate solutions – Bounds
Traffic Grooming for Ring Networks: Heuristics
Heuristic Arbitrary
UPSR/BLSR Mustafa & Kamal ’03
SA Arbitrary
BLSR Wang et al ‘01
GA Arbitrary
UPSR Xu et al
Heuristic Arbitrary
UPSR/BLSR Zhang and Qiao ‘00
Heuristic Arbitrary
BLSR Wan et al ‘00
10/9 approximation Arbitrary
BLSR, single hub
Li et al ‘00
SA Uniform all-to-all Hubbed, and
single hop Cho et al ‘01
Heuristic Uniform all-to-all
BLSR Chiu and Modiano ’00
Heuristic Uniform all-to-all
BLSR Simmons et al ‘98
Result Traffic