a graph-theoretic qos-aware vulnerability assessment for network topologies
TRANSCRIPT
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
A Graph-theoretic QoS-aware Vulnerability
Assessment for Network Topologies
Ying Xuan
March 24, 2010
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
Table of contents
1 Problem Definition and Contributions
2 Discussion on Hardness
3 Algorithms for different network settings
network with small amount of constraints
small networks with unrestricted constraints
two fast heuristic algorithms for general networks
4 Simulation
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
System Model
given a networks with a set of s − t paths;
each path satisfies some of the constraints, thus has a satisfactory score by
assigning each constraint a score.
as long as there exists a path with score higher the some threshold, the network
is still operational.
goal: at least home many node/link failures will make the network not
operational, i.e. no such satisfiable path exists, and QoS routing service is not
available from s to t.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
Graph Model
directed graph G(V , E), source s, destination t;
additive weight vector for each edge ej = (u, v) ∈ E :(
wj1,w
j2, · · · ,w
jm
)
.
weight vector for each path:(
∑
ej∈p wj1,
∑
ej∈p wj2, · · · ,
∑
ej∈p wjm
)
;
constraint threshold vector (c1, c2, · · · , cm);
priority vector (λ1, λ2, · · · , λm);
we define a satisfactory score φ(p) for path p as:
φ(p) =∑
j :p∝i
λj
where p ∝ i iff wpi≤ ci
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
Graph Model
E
H
(2,2)
D
G
(1,1)
F
I(2,2) (3,2)
(2,1)(2,2)
(2,2) (1,1) (2,3)
(1,2)
(3,7)(1,2) BA C
Given a score threshold ρ, find the minimum number of edges whose removal makes
φ(P) = max{φ(p) | ∀s − t path p in G} ≤ ρ
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
Integer Program
Variables:
Xe = 1 if edge e is NOT removed in the optimal solution, 0 otherwise;
Ypi = 1 if a s − t path p has wpi≤ ci , 0 otherwise;
a large constant ǫ = maxi{∑
e∈E wei }.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
Integer Program
min∑
e∈E
(1− Xe)
s.t.∑m
i=1 Ypiλi ≤ ρ, ∀s − t path p∑
e∈p wei≤ ǫ− Ypi ǫ +
∑
e∈p(1− Xe)ǫ + ci , ∀i ,p∑
e∈p wei > 1− Ypi ǫ−
∑
e∈p(1 − Xe )ǫ + ci , ∀i , p
Xe ∈ [0, 1]
Ypi ∈ [0, 1]
Figure: Integer Programming Formulation
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
Main Contributions
provide the first graph-theoretic QoS-ware vulnerability assessment method;
abstract the assessment problem as a graph optimization problem and study its
hardness;
present exact solutions to the problem in two practical cases and several
heuristics for general cases.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
decide version
Given a set of edges deleted, does there exist a path P satisfying φ(P) ≤ ρ in the
remaining graph.
Definition (QoS-SP)
Given a graph with an m-dimension constraint vector, find a path P such that
φ(P) = maxp=[s,··· ,t]∈G
φ(p)
Definition (MCP)
Given m constraints Ci , the problem is to find a path P from a source node s to a
destination node t such that
wi (P) ,∑
(u,v)∈P
wi (u, v) ≤ Ci
for all i ∈ [1, m]Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
NP-hard
Lemma
QoS-SP problem is NP-Complete.
Proof.
Consider the decision version of QoS-SP: given a positive number k, decide if there
exists a path P with φ(P) ≥ k. This proof is straightforwardly by reduction from MCP
problem by letting k =∑m
i=1 λi . Then if there exists a solution for the QoS-SP
problem, then the path is a feasible path to MCP, otherwise, MCP has no solution.
Since MCP is NP-hard, the proof completes.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Exact Solution
Enumerate all (up to 2m) satisfiable constraint states;
Employ FringeMCSP [5] to find the shortest path that satisfies a specific set of
constraints;
Use revised Edmonds-Karp algorithm to remove all satisfiable augmenting path,
w.r.t a set of constraints.
Find a minimum set of edge cut.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Exact Solution
Exact Algorithm
S ← ∅;
for any maximal subset ss of M with∑
ci∈ss λi < ρ do
S ← S ∪ {M \ ss}
while S 6= ∅ do
ss ← extracted from S;
for each edge (i , j) ∈ E do
Set f (i , j) = f (j , i) = 0; Set cf (i , j) = 1 and cf (j , i) = 0;
while shortest path q that satisfies all constraints in ss can be found using
FringeMCSP do
for each edge (u, v) ∈ q do
cf (q) = min{cf (u, v) : (u, v) ∈ q}; f (u, v) = f (u, v) + cf (q);
f (v , u) = −f (u, v); cf (u, v) = c(u, v)− f (u, v);
cf (v , u) = c(v , u)− f (v , u);
all the vertices reachable from s on the residual network induces a cut T .Return the minimum cut among the cuts derived.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Exact Solution
Enumeration of satisfiable constraint states does not work.
Exact Solution:
Employ an existing all-path dynamic programming algorithm label-correcting [3]
to discovery satisfiable paths;
Use revised Edmonds-Karp algorithm [6] to return the cut.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Label-Correcting
Label-Correcting to find satisfiable paths
S ← ∅; labelSet(s) ← {(0, 0, · · · , 0, ∅)};
Q ← {s};
while Q 6= ∅ do
u ← extracted from the end of Q;
for all outgoing edges (u, v) do
labelList′
(v)← Merge(labelList(v), labelList(u), (u, v));
/*extend tuples of u to v by adding the weights of (u, v), eliminate all
dominated tuples.*/
if labelList′
(v) 6= labelList(v) then
labelList(v)← labelList′
(v), Q ← Q ∪ {v};S ← all the PO paths obtained from labelList(t);
Return q = maxp{φ(p) | p ∈ S}.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Strategies
Adopt a Relax-SAT test for estimate the existence of satisfiable path in the
remaining graph;
Adopt two greedy metrics to iteratively remove edges until the Relax-SAT test
returns Negative result.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Relax-SAT test
Relax-SAT Metric
For each edge e,
ϕ1(e) = −m
∑
i=1
wei
ci
λi
For each path p,
ϕ1(p) =∑
e∈p
ϕ(e)
Similarly, define
ϕ2(e) =m
∑
i=1
wei
ci
λi
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
How far from SAT
Assume λ ,∑m
i λi , β(p) , maxi (wp
ci);
Assume p satisfies a set Cs of constrains, and does not satisfy the set Cs , then
φ(p) =∑
ci∈Csλi and ϕ1(p) = −
∑
ci∈Cs
wpi
ciλi −
∑
cj∈Cs
wpj
cjλj , therefore
φ(p) ≥ ρ ⇒ ϕ1(p) ≥ −ρ − (λ − ρ)β(p);
ϕ1(p) ≥ −ρ ⇒ φ(p) ≥ ρ;
since it is hard to calculate β(p) when p is not determinate, we assert
maxϕ1(p) = −minϕ2(p) ≤ −ρ ⇒ maxφ(p) ≤ ρ;
−minϕ2(p) ≥ −ρ ⇒ maxφ(p) ≥ ρ;
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Relax-SAT test
Relax-SAT test
Input: directed graph G = (V , E), constant ρ;
Output: is there a satisfiable path.
for every edge e ∈ E do
ϕ2(e)←∑m
i=1we
ici
λi ;
q ← shortest s-t path on metric ϕ2;
if ϕ2(q) > ρ then
Return NO;
if ϕ2(q) < ρ then
Return YES;
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Two greedy metrics
Definition (Nonlinear Mixed Edge Metric)
ξ(e) =n
∑
i=1
(
wi (e)π(i , e)
ci
)
α
λi
where wi (e) is the i th weight on edge e, π(i , e) is the length of a s − t path, which
has the minimum weight w.r.t the i th constraint over all s − t paths containing e, and
λi is the priority parameter for the i th constraint.
It is evident that the smaller ξ(e) is, the more likely it will be included in the
QoS-optimal path. α ∈ [1, · · · , +∞) is tuned to increase the likelihood of this path to
be selected in the optimal solution.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
network with small amount of constraintssmall networks with unrestricted constraintstwo fast heuristic algorithms for general networks
Two greedy metrics
Definition (Betweenness)
θ(e) ,∑
i : e∈Pi
λi
where Pi refer to the shortest path w.r.t the i th constraint.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
Simulation Results
To be announced.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw
Problem Definition and ContributionsDiscussion on Hardness
Algorithms for different network settingsSimulation
Bibliography
Fernado A. Kuipers and Piet F. A. Van Mieghem, “Conditions that Impact the
complexity of QoS Routing”, IEEE transaction on Networking, 2005
Line B. Reinhardt and David Pisinger, “Multi-Objective and Multi-Constrained
Non-Additive Shortest Path Problems”, Technique Report, DTU Management ,
2009.
Turgay Korkmaz and Marwan Krunz, “Multi-Constrained Optimal Path
Selection”, Infocom, 2001.
P. Kh adivi, S. Samavi and T. D. Todd, “Multi-constraint QoS routing using a
new single mixed metrics”, Journal of Network and Computer Applications, 2008.
Yuxi Li, Janelee Harms, Robert Holte, “Fast Exact MultiConstraint Shortest
Path Algorithms”, ICC, 2007.
Jack Edmonds and Richard M. Karp, “Theoretical improvements in algorithmic
efficiency for network flow problems”, Journal of the ACM 19 (2): 248õ264.
Ying Xuan A Graph-theoretic QoS-aware Vulnerability Assessment for Netw