Optimization of IP Networks in Various Hybrid IGP/MPLS Routing Schemes

Communication Networks
E. Mulyana, U. Killat
1
MMB & PGTS 2004 – Dresden – 14.09.2004
Optimization of IP Networks in Various
Hybrid IGP/MPLS Routing Schemes
Eueung Mulyana, Ulrich Killat
FSP 4-06 Communication Networks
Hamburg University of Technology (TUHH)

2
Models for Hybrid Routing
1
2 3
4 5
6
7
8 9
LSP
1
2 3
4 5
6
7
8 9
LSP
1
2 3
4 5
6
7
8 9
LSP
1 3
2 4
5 6
7 8
9
1
1
1
1
1
1
1
2
2
3
2
2
LSP
Basic IGP Shortcut (BIS)
[2] ...
IGP Shortcut (IS)
[2,7,9][] ...
Overlay (OV)
[5,6] ...

3
Problem Setting
Optimization
Network topology
and link capacities
Traffic demand
Set of LSPs
to be installed
Set of metric
values
Objective(s)
Constraint(s)
 Use single-path routing scenario
 Enforcement of single-path routing :
 ECMP disabled ; or
 Unique shortest path routing [2,8] ...
State of the
network (load
distribution,
etc.)
Set of possible node pairs
for the LSPs (optional)

4
Utilization Upper boundObjective Function
Formulation
}||{min max1
 c max,
 ji Aji  ),(
1|| max
 hN kLSP
k
Hop/Static-Delay limit
 Intended to Heuristics (it is not suitable for Mathematical
Programming)
 Applicable for most core networks
Utilization
 
k
LSP
ji
uv
vu
jiji
k
lll ,
,
,,
ji
ji
ji
c
l
,
,
,

Aji  ),(
max
 kLSP
k

5
A Heuristic Approach : GA
Crossover
Mutation
Population
50 chromosomes
Selection (parents)
8 chromosomes
Selection
(remove 10%)
Population
45 chromosomes
Offsprings
16 chromosomes
Population
61 chromosomes
Selection
(best 50 chromosomes)
Initialize population
Exit condition
fulfilled ?
Parents selection
Crossover
Mutation
Remove some bad individuals
Add new individuals
Survivors selection
An example of the population dynamic
yes
no

6
Representation
2 1 0 2 3 0
LSP A
LSP B
LSP C
LSP D
LSP E
LSP F
LSP G
LSP H
LSP I
LSP J
LSP K
LSP L
LSP M
LSP N
LSP O
LSP P
LSP Q
LSP R
1
2
3
4
Shortest Path
(SP)
4 3 2 2 4 3
4 LSPs are installed
in the network
Possible
LSPs
chromosome
vector of cl


 

else
||if||
k
kPP
c
ll
l
number/
index
 LSP candidates (for each of the considered flows) are obtained by
applying k-shortest path algorithm
4|| 

7G-WiN Level 2G-WiN Level 1The G-WiN Network
Case Study
Network
instance
27 nodes (10 level-1 + 17 level-2);
76 directed-links (42 level-1 + 34 level-2)
Static delay
model cspproplink
 
prop

sp

c

propagation
proc./serial.
constant
Parameter
setting
(622Mbps)ms1sp

linkperms1c

ms12max 
4max
h
10001
c
cdprop  2
3


8
Results(1)
#LSP
IGP
BIS IS
0.7725 0.570338
0.7725
cons
max

0.2345
-
58
link
max

24.9474
link

-
LSP
max

-
LSP

OV
using set of the
considered links,
in this case all level-1
links
cons
max

max

maximum #different flows
carried by an LSP
link
max

average #different flows
carried by a link
link

average #different flows
carried by an LSP
maximum #different flows
carried by a link
LSP
max

LSP

IGP/MPLS
0.570338
0.237114
12
53
25.0526
1
1
0.523714
0.523714
0.240888
16
45
25.4079
4
2.9375
0.483923
0.426447
0.247998
22
51
26.5789
12
8.54545
max
Termination : 150 iterations (maximum) or 70 iterations (no improvements)
GA : population size = 30;
IGP : inverse-capacity metric values; ECMP was disabled
L=48 (all node pairs in the level-1 network, which are not directly connected)
path)shortest-(k4k

9
Results(2) : Delay
IGP OV BIS IS
path
 6.687 6.738 6.904 7.238
)(flows#  - 1.42% 5.7% 21.08%
)(flows#  - 0.28% 1% 5.7%
)(flows#  - 98.3% 93.3% 73.22%
path
max
 12.1697 12.1697 13.1659 14.8418
- 11.1755 10.7706 10.3724
LSP
max

- 8.30118 8.72935 7.56434
LSP


10
Results(3) : Hop-count
IGP OV BIS IS
path
h 2.70 2.71 2.75 2.88
)(flows# h - 1.14% 4.99% 17.66%
)(flows# h - 0% 0% 0%
)(flows# h - 98.86% 95.01% 82.34%
path
hmax
4 4 4 5
- 3 3 3
LSP
hmax
- 2.67 2.75 2.64
LSP
h

11
Summary and Conclusion
 Investigation of various hybrid IGP/MPLS routing schemes related to
network performance and QoS parameters
 Hybrid IGP/MPLS attractive alternative and complement for
traditional offline TE by optimizing IGP metrics.
 New approach based on GA; increasing network efficiency while
minimizing the necessary number of LSPs within certain performance
trade-offs
 Advantages vs. disadvantages:
 OV: no-aggregation, relatively large number of LSPs are needed
to achieve a good performance
 BIS+IS: aggregation is possible, relatively smaller number of
LSPs are –if appropriately configured– enough to achieve a good
performance

12
References (Partial List)
(1) Awduche D. et. al. „Overview and Principles of Internet Traffic Engineering“, RFC
3272, May 2002.
(2) Ben-Ameur W. et. al. „Routing Strategies for IP-Networks“, Telekronikk Magazine
2/3, 2001.
(3) Forzt B., Thorup M. „Internet Traffic Engineering by Optimizing OSPF Weights“,
IEEE Infocom, 2000.
(4) Karas P., Pioro M. „Optimisation Problems Related to the Assignment of
Administrative Weights in the IP Networks‘ Routing Protocols“, Proceedings of 1st
PGTS 2000.
(5) Koehler S., Binzenhoefer A. „MPLS Traffic Engineering in OSPF Networks – A
Combined Approach“,ITC 18, 2003.
(6) Riedl A. „Optimized Routing Adaptation in IP Networks Utilizing OSPF and MPLS“,
IEEE ICC, 2003.
(7) Shen N., Smit H.„Calculating IGP Routes over TE Tunnels“, Internet Draft, 1999.
(8) Thorup M., Roughan M. „Avoiding Ties in Shortest Path First Routing“,[online].
(9) Wang Y., Zhang L., „A Scalable Hybrid IP Network TE Approach“, Internet Draft,
2001.

13
Thank You !

14
Queueing Delay
 [Papagiannaki_02] „Anaysis of Measured Single-Hop Delay from
an Operational Backbone Network“ (Tier-1/Sprint) :
 „99% of packets in the backbone experience single-hop delay
less than 1 ms“ (max 35 ms; maximum link utilization of
70%; average link utilization of 50%; investigation on OC-
3/STM-1 link; GPS-clock)
 Cisco recomendation; [Filsfils_02] „Deploying Tight-SLA Services
on an IP Backbone“:
 „based on simulations for at least 85% utilization with a mean
queue size below 20 ms“(using the recommended WRED
parameter)

15
Crossover and Mutation
2 1 0 2 3 0
1 0 1 2 2 3
4 1 2 0 1 1
1 0 1 1 0 1
2
1
0 2
3
0
1
0
1 2
2
3
0 0 0 1 0 0
4 1 2 2 1 1
Chr. 1
Chr. 2
Crossover
vector
Chr. 3
Chr. 4
Chr. 5
Chr. 6
Mutation
vector
4 3 2 2 4 3Vector of cmax

16
Applicability for Using max
 Should be ensured that it is always possible to reroute load in each
link in the network
 If not the case : transform
.
maxmax
cons
 
2
3
4 5
6
7
1
considered
not
considered
Mark each link as
considered or
unconsidered

17
|Rall|  cardinality of all routing entries for all nodes
|D|  cardinality of all demand entries
c (>>)  a quite high constant
Objective function
Unique Shortest-Path Routing :
Heuristics
 Thorup, Roughan „Avoiding Ties in Shortest Path First Routing“
 Using an explicit penalty in the objective function :
2
3
41
vu
f ,
2
,vu
f
2
,vu
f
2
3
41



c
fcf vuvu ,,
2
,

vu
f
2
,

vu
f
|)||(| all
DRc 



18
The Normalized Objective Function
Utilization Upper boundObjective Function
}||
1
{min max1

L
c  max,
 ji Aji  ),(
1|| max
 hN kLSP
k
Hop limitUtilization
 
k
LSP
ji
uv
vu
jiji
k
lll ,
,
,,
ji
ji
ji
c
l
,
,
,

Aji  ),(
 L  length of the chromosome = max. LSPs can be installed (cf. GA
representation)
max
 kLSP
k

19
IGP Shortcut : Routing Loop
1
2 3
4
5
6
7
8 9
LSP
1 3
2 4
5 6
7 8
9
1
1
1
1
1
1
1
2
2
3
2
2
LSP
1 2 4 6 8 7 5 7
8
9
LSP
Routing
Loop
 Bad configuration! It is certainly not
optimal w.r.t. network utilization
 Using an explicit penalty in the
objective function (cf. Page 18), if
there exist routing entries which
contain loop

20
IGP Shortcut : The Algorithm [7]
1 2 3 4 5 6 7 8 9
1 2 3 4 5 6 7 8 9
LSP B
LSP A
LSP C
LSP D
LSP A
LSP B
LSP C
LSP D
LSP E

21
Convergence (1)
Iterations
Objective value
(fitness)
fitness = 443.987
utilMax = 0.483923
utilMaxConsidered = 0.413987
totalTunnels = 30
---------------------------
fitness (bestIndividual) = 443.987
utilMaxCons (b.I.) = 0.413987
totalTunnels (b.I.) = 30
----------------------------
Total Computation Time = 32.3 minutes
Computation Time (Network) = 32.25 minutes
Computation Time (Network) = 99.8452 %
----------------------------
Average
Best Individual

22
Convergence (2)
Iterations
Maximum utilization
(considered) # LSPs
Average
Best Individual
Best Individual
Average

Optimization of IP Networks in Various Hybrid IGP/MPLS Routing Schemes

More Related Content

What's hot (16)

Viewers also liked (13)

Similar to Optimization of IP Networks in Various Hybrid IGP/MPLS Routing Schemes (20)

Recently uploaded (20)

Optimization of IP Networks in Various Hybrid IGP/MPLS Routing Schemes