Transcript presentation - University of Alberta

Quantitative Comparison of End-to-End
Availability of Service Paths in Ring and MeshRestorable Networks
Matthieu Clouqueur, Wayne D. Grover
[email protected], [email protected]
National Fiber Optic Engineers Conference - NFOEC 2003
Orlando, Florida, USA
Outline
• Motivation
• Goals of Study
• Survivability Schemes Models
• Description of Test Cases
• Sample Results
• Discussion of Results
• Conclusion
2
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Motivation
• Rings are often associated with high availability because they provide
high restoration speed.
• Mesh is often thought to provide lower availability associated with low
capacity redundancy and slower restoration.
We ask:
• Which really provides higher availability: ring or mesh ?
• How do ring and mesh compare in service path availability and in the
range of availability levels they can offer ?
• In mesh, what would be the effects of affecting priorities to selected
services ?
3
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Goals and Methodology
• Provide a true “apples-to-apples” comparison of end-to-end service
path availability in ring and mesh:
– Ring and Mesh are compared on identical facilities graphs serving identical
demands
– Efficient capacity designs are used for both architectures
– Exact survivability mechanisms are emulated
– Both architectures experience identical failure sequences
• Comparison is based on:
– Average path unavailability (versus path length)
– Average number of outages experienced per year per service path (versus
path length)
– Statistical frequency of total path outage times per year
4
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Simulation Numerical Details
• Statistics of Failures:
– Mean time between failures (MTBF): 1 year on each span
– Negative exponential distribution (Poisson process)
• Statistics of Repair:
– Mean time to repair (MTTR): 12 hours
– Negative exponential distribution
5
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Main Assumptions of the Availability Analysis
• Previous work shows that what dominates service unavailability is:
–
–
–
–
Reconfiguration times (to single failures)
Single node-failures
Dual span-failures
Triple failures
• The present analysis is based on the effects of dual span-failures
– The contribution of reconfiguration times is not taken into account.
• The analysis is based on restorability investigations performed on a per
path basis for 7each particular failure scenario occurring in the
simulation
6
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Modelling Ring Survivability
• The Bi-directional Line Switched Ring (BLSR) model is assumed:
surviving
path
Working channels
1
Loop-back
5
Note: A dual fibre cut affecting a ring
does not cause all service paths on
the ring to experience outage
 Detailed inspection is performed
on a per path basis to identify paths
affected by the dual failure.
2
Protection channels
3
4
Loop-back
failed
path
7
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Modelling Mesh Survivability
• Adaptive mesh restoration behaviour is assumed:
– Restoration paths for a failed span are dynamically searched within
remaining spare capacity (not according to pre-plan)
7
Note: Restoration of a failed span
includes an effort to restore any
spare capacity used on that span
 A restoration path affected by
a second failure may survive
6
23
21
24
10
9
15
22
20
11
12
18
19
14
13
17
16
Restoration to a first failure can
be based on a pre-plan.
Second failure response is
adaptive
failed backup path
is itself restored
8
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Mesh Restoration with Priorities
• Capacity designs used are identical to normal single failure restorable
designs (no additional capacity)
• A certain percentage of demands is given “Priority” status on each
origin/destination node pair
– The Priority service paths are considered first for restoration and will
therefore have a higher dual-failure restorability and therefore a higher
availability
• Three service mixes: 10/90, 30/70, 50/50 (% high P. / % low P.)
Questions:
– How much is the availability of Priority services improved?
– How much is the availability of non-Priority services degraded?
9
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Three test cases
Test case
net32-A
net32-B
25n50s1
N04
29
N02
N03
30
N01
26
27
N05
N25
22
1
28
N08
21
N07
2
10
3
N06
16
11
12
N23
15
N21
4
N24
N10
N09
N15
32
14
Topology
17
N22
N20
24
31
N12
13
N11
N18
N19
8
6
5
N13
7
9
23
20
19
(32 nodes, 45 edges)
Type of
Demand
N14
18
25
Hubbed Demand
Matrix
N17
N16
(25 nodes, 50 edges)
Gravity-based
Demand Matrix
10
Gravity-based
Demand Matrix
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Study Details
• Mesh:
– Working paths are routed on shortest path
– Minimal spare capacity placed by Integer Linear Programming Optimization*
– Average working path length: net32-A: 5.5; net32-B: 2.2; 25n50s1: 2.8
• Ring:
–
–
–
–
Ring designs using efficient methods developed in PhD work by D. Morley**
# of rings in design: net32-A & net-32B: 8 OC-48; 25n50s1: 19 OC-48
Average number of spans/ring: net32-A & net32-B: 11; 25n50s1: 9.4
Working paths routed by shortest ring-constrained routing
* J. Doucette, W. D. Grover, “Influence of modularity and economy-of-scale effects on design of mesh-restorable DWDM networks,” IEEE JSAC,
vol. 18, no. 10, October 2000, pp. 1912-1923.
** G. D. Morley, Analysis and Design of Ring-Based Networks, Ph.D. Thesis, University of Alberta, Spring 2001.
and G. D. Morley, W. D. Grover, “Tabu search optimization of optical ring transport networks,” in Proceedings of IEEE GLOBECOM 2001,
November 2001, vol. 4, pp. 2160-2164.
11
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Statistical Considerations
• Results for each test case are based on series of 1000 one-year
simulations
• Total of dual (or higher order) failures arising over 1000 trials:
– Net32-A & Net32-B: 2619
– 25n50s1: 3180
• Average number of outage events per path being the basis for
availability results:
– Ring: 64.7
– Mesh: 45.9
• This was shown to give good confidence levels on results:
– E.g. for 25n50s1 test case, over 10 separate 1000 one-year trials, standard
deviation of average unavailability Uave results was calculated to be 2.7 %
of Uave
12
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Comparative Results of Path Unavailability
Average working path lengths
2.0E-04
net32-A Ring
net32-B Ring
25n50s1Ring
net32-A M esh
net32-B M esh
25n50s1M esh
1.8E-04
Path unavailability
1.6E-04
1.4E-04
1.2E-04
~ 26 % chance of
outage in a given
year (worst case)
net32-A
25n50s1
1.0E-04
~ 17 % chance of
outage in a given
year (worst case)
net32-B
8.0E-05
6.0E-05
4.0E-05
2.0E-05
0.0E+00
0
2
4
6
Number of hops in path
8
10
Path unavailability is significantly higher in the ring architecture, especially for
longer paths (up to a factor 2 in the worst case)
13
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Comparing Distributions of Outage Times
Results for test case 25n50s1
90th
median percentile
~ 6 hours 15 hours
0.6%
25n50s1 Ring
0.5%
Frequency
Ring: 93.7 %
of paths
experience no
outage in a
year
0.4%
w orst case: 72 hours
(Prob = 2.5x10-4 %)
0.3%
0.2%
0.1%
0.0%
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
Total outage per year (hours)
0.6%
25n50s1 Mesh
0.5%
Frequency
Mesh: 95.4 %
of paths
experience no
outage in a
year
90th
median
percentile
~ 6 hours 13.5 hours
0.4%
0.3%
w orst case: 48 hours
(Prob = 6x10-5 %)
0.2%
0.1%
0.0%
0
5
10
15
20
25
30
35
40
45
50
55
60
65
70
Total outage per year (hours)
14
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Effects of Priorities in Mesh
1.0E-04
Results for test case 25n50s1
Ring
M esh
M esh Low P . (50/50)
M esh Low P . (70/30)
M esh Low P . (90/10)
M esh High P . (50/50)
M esh High P . (70/30)
M esh High P . (90/10)
Path unavailability
8.0E-05
6.0E-05
The availability of
non-Priority services
remains comparable
or better than in ring
A priority path in
mesh can have a
more than five times
lower unavailability
than in ring
4.0E-05
2.0E-05
0.0E+00
0
1
2
3
4
Number of hops in path
5
6
7
The effect of prioritizing in mesh is a considerable reduction of
path unavailability for the Priority class.
The non-Priority service class suffers only a small degradation of
service availability.
15
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Additional Insights from the Study
• The effect of priorities is a significant reduction of the probability of
paths experiencing outage in a given year:
– E.g. for 6-hop paths in the 25n50s1 test case, the probability of
experiencing outage in a given year drops from 12% to 2% is the path is
included in the 10% priority class (10/90 scheme).
• Experimental results show that the advantage of mesh over ring is
greater in the highly connected topology
– This is especially true for results of prioritized mesh and confirms the
understanding that mesh benefits greatly from high network diversity
• Under the 10/90 service mix Priority services achieve the lowest
unavailability while the availability of non-Priority services remains
virtually unchanged
16
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003
Conclusion
• Despite its lower capacity requirements, the mesh architecture
achieves better availability than ring
– The key is mesh’s better ability to withstand dual-failure states even with
less redundancy than rings.
• With prioritization of services, high priority services can be offered very
low unavailability (3 to 5 times less than with rings) while non-Priority
service still enjoy comparable or better availability than with rings.
Concluding comment:
What matters most for high availability is not fast restoration to
single failures but the ability to provide high restorability to dual
failures. This minimizes the probability of MTTR-scale outages on
service paths.
17
Matthieu Clouqueur and Wayne D. Grover
NFOEC 2003