routing optimization in ip/mpls networks under per-class over-provisioning constraints
TRANSCRIPT
Communication Networks E. Mulyana, U. Killat
1
INOC 2005 – Lisbon – 22.03.2005
Routing Optimization in IP/MPLS Networks under Per-Class
Over-Provisioning Constraints
Eueung Mulyana, Ulrich Killat FSP 4-06 Communication Networks
Hamburg University of Technology (TUHH)
Communication Networks E. Mulyana, U. Killat
2
Hybrid Intra-Domain IP Routing
Vanilla LSP
ER LSP
2
1 2
3 5
2
5
1 2
3 4
5 6
Link Weights
1
2 3
4 5
6
1 2
3 4
5 6
MPLS allows explicit (using ER-LSPs) other than shortest path routing (using Vanilla LSPs)
DiffServ gives possibility to differentiate treatements for IP packets with respect to their class of service e.g. class-based routing
Communication Networks E. Mulyana, U. Killat
3
Over-Provisioning (OP) (1)
Avoiding overload by ensuring that capacity of all links is greater than demand both in normal or in failure situations
large variations in traffic demand
QoS: the service that traffic receives is dependent upon the offered load and the available capacity
Simple capacity provisioning rules: upgrade when utilization reaches 40-50% over-provisioned by a factor of 2
Communication Networks E. Mulyana, U. Killat
4
Over-Provisioning (2)
VoIP 150 Mbps
best-effort(BE) 1.5 Gbps
Aggregate OP 2 lines of 2.5 Gbps
Per-class OP 1 line of 2.5 Gbps
Aggregate vs. per-class over-provisioning
2)Aggr(OP
c 03.3)Aggr(
4)VOIP(OP
c
2.1)BE(OP
c
67.16)VOIP(
57.1)BE(
t)(constrain valueOPgiven OP
c
valueOP actual
Per-class OP approach offers better service (guarantee) for prioritized VoIP class without large over-provisioning of capacity!
Communication Networks E. Mulyana, U. Killat
5
Over-Provisioning Factor
1
1
,,
*
,
s
s
jijiji lcc
jiji cc ,
1*
,
2*
, jic
1
, jil
2
, jil
jic ,
1
, jil
52
101
, ji
24
82
, ji
jic ,
1
, jil
2
, jil
52
101
, ji 67.142
102
,
ji
ji
ji
ji
l
c
,
*
,
,
1
,
,
,
s
s
ji
ji
ji
l
c
per-class cumulative
Communication Networks E. Mulyana, U. Killat
6
Problem Setting
Set of metric values (unique shortest path) for establishing vanilla Label Switched Paths (LSPs)
Set of Explicit Route (ER)-LSPs
Capacitated network
Traffic matrices of different classes
Per-class OP constraints
Per-class load distribution & per-class OP profile
Optimization
OP
,
*
,
),(min }{min
c
l
c
ji
ji
Aji
Actual minimum OP value
Given minimum OP constraint
Communication Networks E. Mulyana, U. Killat
7
Label Switched Path (LSP) Design (1) Indirectly solved by iteratively calling a metric-based traffic
engineering (TE) procedure using traffic matrices of different classes
F aggregate traffic matrix Fi traffic matrix for class i RT base routing pattern (obtained via optimization using F ) RTi routing pattern for class i (obtained via optimization using Fi)
Communication Networks E. Mulyana, U. Killat
8
Label Switched Path Design (2)
optimize network(F)
optimize network(F1)
optimize network(F2)
optimize network(F3)
s
s
3
1
}3,2,1{
Weight System (WS)
base (WS0)
WS1
WS2
WS3
-
1
2
-
-
3
-
-
-
ji
ji
Aji c
l
,
,
),(max max ||
An example:
Objectives:
Minimizing and
Communication Networks E. Mulyana, U. Killat
9
Simulated Annealing Approach for Optimization Task (1)
Utilization Upperbound
Objective Function
} { min*
max
Al
lyc
*
max
*
, ji Aji ),(
Utilization
uv
uv
jiji ll )(,,
*
,
,*
,
ji
ji
ji
c
l Aji ),(
otherwise0
1 ww
y
o
ll
l
RTwwwwo
A
o
l
oo||21 ,,,,,
iAl RTwwww ||21 ,,,,,
Reference weight system:
A solution (current weight system):
LSPsER RT
Communication Networks E. Mulyana, U. Killat
10
Representation
Move Operator
Simulated Annealing Approach for Optimization Task (2)
2 1
3 4
5 6
w1
2 1 2 2 3 5 5
21 w
12 w
35 w
23 w
24 w
56 w
57 w
w2 w3 w4 w5 w6 w7
Simulated Annealing
w1
2 1 2 2 3 5 5
w2 w3 w4 w5 w6 w7 w1
2 1 5 2 3 5 5
w2 w3 w4 w5 w6 w7
Communication Networks E. Mulyana, U. Killat
11
Simulated Annealing Approach for Optimization Task (3)
Joint with plain local search (PLS):
concentrate the search around best solution (small ) first before exploiting other regions
speed-up convergence at small number of iterations
0
))()'(
exp(
1
T
xxp
)()'( xx
0 and )()'( PLS xx
otherwise
otherwise
satisfied are PLS performingfor conditions if
0
1
PLS
Communication Networks E. Mulyana, U. Killat
12
Case Study
3
13
9
14
11
8
10
6
5
74
2
1
12
2500 Mbps
net14 #nodes #links
14 nodes 44 links (directed)
effort)-(best 3
(assured) 2
(premium) 1
demands
interval mean
6.68 3
0.73 2
49.6 1
]556,0[ 3
]70,10[ 2
][10,150 1
Communication Networks E. Mulyana, U. Killat
13
Results: net14 (1)
0.4OP
1 c
0.4OP
2 c After optimize network(F)
i.e. without ER-LSPs:
)1.1|4.3|3(min
%44.96max
Communication Networks E. Mulyana, U. Killat
14
Results: net14 (2)
0.4OP
1 c
0.4OP
2 c
After optimize network(F2) : 13 symmetrical ER-LSPs (premium) and 4 symmetrical ER-LSPs (assured)
)1.1|01.4|05.4(min
%68.93max
Communication Networks E. Mulyana, U. Killat
15
Summary and Conclusion
Study of offline routing control in multi-class IP/MPLS networks:
using hybrid routing scheme
taking per-class OP constraints into account
Proposing a simple heuristic which iteratively calls a metric-based TE procedure, that minimizes per-class maximum utilization while minimizing the number of ER-LSPs
Starting from an optimized weight system for traffic aggregates, a few ER-LSPs are installed to improve minimum OP factors for each class
Communication Networks E. Mulyana, U. Killat
16
Thank You !
Communication Networks E. Mulyana, U. Killat
17
References (Partial List) (1) Filsfils C., Evans J. „Engineering a Multiservice IP Backbone to
Support Tight SLAs“, Int. J. of Computers and Telecommunications Networking 40/1:131-148, 2002.
(2) Ben-Ameur W. et. al. „Routing Strategies for IP-Networks“, Telekronikk Magazine 2/3, 2001.
(3) Fortz B., Thorup M. „Internet Traffic Engineering by Optimizing OSPF Weights“, Proc. IEEE Infocom, 2000.
(4) Blake S. et al. „An Architecture for Differentiated Services“, RFC 2475, 1998.
(5) Le Faucher F. et al. „MPLS Support of Differentiated Services“, RFC 3270, 2002.
(6) Roberts J.W. „Traffic Theory and the Internet“, IEEE Communication Magazine, 2001.
(7) Smith P.A., Jamoussi B. „MPLS Tutorial and Operational Experiences“, NANOG 17 Meeting, 1999.
Communication Networks E. Mulyana, U. Killat
18
Calculating Link Load and Utilization
uv
uv
jiji ll )(,,
k
k
ji
uv
uv
jiji lll ,,, )(
jiji ll ,,
Link load for class
Link load for aggregate traffic
Using WS : Using WS0 :
only exist if (u,v) are not (head,tail) nodes of LSP in
Utilization
*
,
,*
,
ji
ji
ji
c
l
ji
ji
ji
c
l
,
,
,
ji
ji
ji
c
l
,
,
,
per-class utilization aggregate utilization effective per-class utilization
Communication Networks E. Mulyana, U. Killat
19
Results: net6
After optimize network(F) i.e. without ER-LSPs
OP for aggregate 1.26
After optimize network(F1) : 1 symmetrical ER-LSP
OP for aggregate 1.19
0.3OP
1 c