Although two monitors are not sufficient to identify all link metrics in G, we explore in this section the case of identifying the entire network using three or more monitors.
3.3.1
Conversion into Two-Monitor Problem
Section3.2suggests that it is easier to identify links that are one-hop away from the monitors. This observation motivates us to construct an extended graph Gex of G.
As illustrated in Fig.3.8, given a graph G with κ monitors, its extended graph Gexis
obtained by adding two virtual monitors m+
1 and m+2, and 2κ virtual links between
each pair of virtual-actual monitors. In this way, all links of interest (actual links in G) are at least one-hop away from virtual monitors m+
(a) (b) ... ... ... ... ... ... mm1 i mj m1 mi mj κ m P m+1 m+2 G Gex G virtual link κ m
Figure 3.8: (a) G with κ (κ ≥ 3) monitors; (b) Gex with two virtual monitors: m+1
and m+ 2.
immediately converts the problem of identifying G using κ monitors to a problem of identifying the interior graph of Gexusing two monitors (again we have no prior
knowledge of link metrics in G or Gex). Therefore, we can apply Theorem 3.3 to
obtain the following result.
Lemma 3.6. Employing κ (κ ≥ 3) monitors to measure simple paths, the necessary
and sufficient condition on the network topology G for identifying all link metrics in G is that the associated extended graph Gex has an identifiable interior graph, i.e.,
Gexsatisfies Conditions and1 in Theorem2 3.3.
Proof. Since G is the interior graph of Gex, it suffices to show that the information
attainable by the real monitors m1, . . . , mκis the same as the information attainable
by the virtual monitors m+
1 and m+2, if the virtual monitors can make end-to-end
measurements along simple paths in Gex.
First, we show that any measurement between the real monitors can be obtained from measurements between m+
1 and m+2. To this end, consider a path miPmj
(i, j ∈ {1, . . . , κ}, i 6= j) in G, as shown in Fig.3.8(b). Four simple paths between m+1 and m+2 can be constructed:
PA= m+1mim+2, PB = m+1mjm+2, PC = m+1miPmjm+2, PD = m+1mjPmim+2. (3.9)
Viewing miPmjas a “cross-link”, we can compute WmiPmj from the measurements
on these four paths via (3.6) (replacing Wy by WmiPmj).
Moreover, we show that measurements between m+
1 and m+2 in Gex do not pro-
vide extra information for identifying links in G compared with measurements at- tainable by the real monitors. This is proved by observing that for any m+
1 → m+2
simple path m+
1miPmjm+2 (i, j ∈ {1, . . . , κ}, i 6= j) containing at least one link in
G, the information relevant for identifying links in G can be obtained by measuring its mi → mj sub-path miPmj, which must also be a simple path.
3.3.2
Complete Identifiability Condition
The special structure of Gex allows us to consolidate the two Conditions and1 2
into a single condition, stated as follows.
Theorem 3.7. Assume that κ (κ ≥ 3) monitors are used to measure simple paths.
The necessary and sufficient condition on the network topology G for identifying all link metrics in G is that the associated extended graph Gexbe 3-vertex-connected.
Proof. To prove Theorem3.7, we first have two claims.
1) Claim A.Given a graph G employing κ (κ ≥ 3) monitors, the extended graph
Gex of G satisfies Conditions (i.e., G1 ex− l is 2-edge-connected for each link l in
G) if and only if Gex is 3-edge-connected.
Claim A is proved as follows:
Necessary part.
Suppose Gex− l is 2-edge-connected for all l in G. Consider removing two links
in Gex, denoted by l1and l2.
(a) If at least one of these links, say l1, is in L(G), then by assumption Gex− l1is
2-edge-connected. Thus, Gex− l1− l2 is connected.
Since the virtual monitors m+
1 and m+2 each connect to all actual monitors in G, and
there are at least 3 actual monitors, m+
1 and m+2 are each connected to G via at least
3 virtual links. Therefore, m+1 and m+
2 are still connected with G after l1 and l2 are
deleted. Since we have assumed G to be a connected graph, Gex−l1−l2is connected.
We have shown that Gexremains connected after removing any two links. There-
fore, Gexis 3-edge-connected when Gex− l (l ∈ L(G)) is 2-edge-connected.
Sufficient part.
Suppose Gex is 3-edge-connected. Then obviously, Gex− l is 2-edge-connected
for each l ∈ L(G).
2) Claim B.Given a graph G employing κ (κ ≥ 3) monitors, the extended graph
Gexof G satisfies Conditions (i.e., G2 ex+m+1m+2 is 3-vertex-connected) if and only
if Gexis 3-vertex-connected.
Claim B is proved as follows:
Necessary part.
We prove the necessary part by contradiction. Suppose Gex is not 3-vertex-
connected, but Gex + m+1m+2 is 3-vertex-connected, then the connectivity of Gex
must be 2, because removing one link will decrease connectivity by at most 1. Thus, there must exist two nodes, denoted by v1 and v2, whose removal will disconnect
Gex. There are 3 possibilities for v1and v2.
(a) If v1, v2 are m+1, m+2, then after their removal, the remaining graph (G) is still
connected.
(b) If v1 is a virtual monitor (m+1 or m+2) and v2 is a node in G, then Gex− v1−
v2 being disconnected will imply Gex + m+1m+2 − v1 − v2 also being disconnected
(as the remaining graphs of Gex and Gex + m+1m+2 are the same), contradicting the
assumption that Gex+ m+1m+2 is 3-vertex-connected.
(c) If v1, v2 are both in G (can be real monitors), then two cases may occur
v1
v2
(a) direct link (b)
m2 m1 m3 m1 m2 + + G1 G2 G1 G2 v1 v2 m1 m2 m3 m+1 m2+ m1+m+2
Figure 3.9: Gex− v1− v2 is disconnected, where v1, v2 ∈ V (G).
real monitor; (b) each connected component contains at least one real monitor, as illustrated in Fig.3.9. In the case of Fig.3.9(a), Gex+m1+m+2−v1−v2is disconnected
as well, contradicting the 3-vertex-connectivity of Gex + m+1m+2. In the case of
Fig. 3.9 (b), different components in Gex − v1 − v2 can still connect via virtual
links and virtual monitors, thus contradicting the assumption that Gex− v1− v2 is
disconnected.
Hence, when Gex+ m+1m+2 is 3-vertex-connected, the connectivity of Gex cannot
be less than 3, i.e., Gex is also 3-vertex-connected.
Sufficient part.
If Gexis 3-vertex-connected, then after adding one link m+1m+2, Gex+ m+1m+2 is
also 3-vertex-connected.
3) From the structure of Gex (see Fig. 3.8), Claims A and B suggest that Gex
satisfies Conditions and1 in Theorem2 3.3 if and only if Gex is both 3-edge-
connected and 3-vertex-connected. According to Proposition 1.4.2 in [47], a 3- vertex-connected graph is also 3-edge-connected. Thus, the necessary and sufficient conditions in Lemma3.6can be simplified to a single condition that Gexbe 3-vertex-