ArticlePDF Available

Grid spanners with low forwarding index for energy efficient networks

Authors:

Abstract and Figures

A routing R of a connected graph G is a collection that contains simple paths connecting every ordered pair of vertices in G. The edge-forwarding index with respect to R (or simply the forwarding index with respect to R) π(G, R) of G is the maximum number of paths in R passing through any edge of G. The forwarding index π(G) of G is the minimum π(G, R) over all routings R’s of G. This parameter has been studied for different graph classes [14], [1], [7], [5]. Motivated by energy efficiency, we look, for different numbers of edges, at the best spanning graphs of a square grid, namely those with a low forwarding index.
Content may be subject to copyright.
HAL Id: hal-01095179
https://hal.inria.fr/hal-01095179
Submitted on 15 Dec 2014
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of sci-
entic research documents, whether they are pub-
lished or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diusion de documents
scientiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Grid spanners with low forwarding index for energy
ecient networks
Frédéric Giroire, Stéphane Pérennes, Issam Tahiri
To cite this version:
Frédéric Giroire, Stéphane Pérennes, Issam Tahiri. Grid spanners with low forwarding index for
energy ecient networks. [Research Report] RR-8643, INRIA Sophia Antipolis; INRIA. 2014. <hal-
01095179>
ISSN 0249-6399 ISRN INRIA/RR--8643--FR+ENG
RESEARCH
REPORT
N° 8643
December 2014
Project-Team COATI
Grid spanners with low
forwarding index for
energy efficient networks
Frédéric Giroire, Stéphane Pérennes , Issam Tahiri
RESEARCH CENTRE
SOPHIA ANTIPOLIS – MÉDITERRANÉE
2004 route des Lucioles - BP 93
06902 Sophia Antipolis Cedex
Grid spanners with low forwarding index for energy
efficient networks
Frédéric Giroire, Stéphane Pérennes , Issam Tahiri
Project-Team COATI
Research Report n° 8643 — December 2014 — 11 pages
Abstract: A routing Rof a connected graph Gis a collection that contains simple paths connecting every
ordered pair of vertices in G. The edge-forwarding index with respect to R(or simply the forwarding index
with respect to R)π(G, R)of Gis the maximum number of paths in Rpassing through any edge of G. The
forwarding index π(G)of Gis the minimum π(G, R)over all routings R’s of G. This parameter has been
studied for different graph classes [14], [1], [7], [5]. Motivated by energy efficiency, we look, for different
numbers of edges, at the best spanning graphs of a square grid, namely those with a low forwarding index.
Key-words: spanning subgraphs, forwarding index, energy saving, routing, grid
This work has been partially supported by ANR project Stint under reference ANR-13-BS02- 0007, ANR program Invest-
ments for the Future under reference ANR-11-LABX-0031-01, ANR VISE, ANR Skyflow, CNRS-FUNCAP project GAIATO, the
associated Inria team AlDyNet, the project ECOS-Sud Chile and Paca Region.
CNRS, Laboratoire I3S, UMR 7172, UNS, CNRS, Inria, COATI, 06900 Sophia Antipolis
Université de Bordeaux, Institut de Mathématiques, UMR 5251, CNRS, Inria, 33405 Talence
Graphes couvrants de la grille avec un indice de transmission faible
pour des réseaux efficaces en énergie
Résumé : Un routage Rd’un graphe connexe Gest un ensemble de chemins simples connectant toutes
les paires ordonnées de sommets de G. Lindice de transmission arête relatif à R(ou simplement l’indice
de transmission relatif à R)π(G, R)de Gest le nombre maximum de chemins de Rqui utilisent une arête.
Lindice de transmission de G,π(G), est le π(G, R)minimum sur tous les routages Rde G. Ce paramètre
a été étudié pour différentes classes de graphes [14], [1], [7], [5]. Motivés par l’efficacité énergétique des
réseaux, nous étudions, pour un nombre d’aêtes donné, les meilleurs sous-graphes couvrants de la grille
carré, c’est-à-dire ceux avec un indice de transmission faible.
Mots-clés : sous-graphes couvrants, indice de transmission, efficacité énergétique, routage, grille
Grid spanners with low forwarding index 3
1 Introduction
A routing Rof a given connected graph Gof order Nis a collection of N(N1) simple paths connecting
every ordered pair of vertices of G. The routing Rinduces on every edge eaload that is the number of
paths going through e. The edge-forwarding index (or simply the forwarding index) π(G, R)of Gwith
respect to Ris the maximum number of paths in Rpassing through any edge of G. It corresponds to the
maximum load over all edges of the graph when Ris used. Therefore, it is important to find routings
minimizing this index. The forwarding index π(G)of Gis the minimum π(G, R)over all routings R’s
of G. This parameter has been studied for different graph classes (examples can be found in [1], [7], [5])
and this survey [14] gives a global view on the known results.
We call a connected spanning subgraph of a graph G, a spanner of G. More precisely, it is a connected
subgraph that has the same set of vertices as G. Our goal is to find, for a given bound on the number
of edges, the best spanner of G, namely the one with the minimum forwarding index. The problem can
also be viewed as: for a given bound Uon the forwarding index, find a spanner Fof Gwith minimum
number of edges such that π(F)U.
Knowing how to solve this problem is very interesting in practice for network operators willing to
reduce the energy consumed by their networks. In fact, most of the network links consume a constant
energy independently of the amount of traffic they are flowing [2], [13]. Therefore, it was proposed to
reduce the energy used by the network links by turning some of them off, or more conveniently, putting
them into an idle mode. Outside the rush hours, several studies [3], [4], [12], [9], [10], [11], show
that a good choice of the links to turn off can lead to significant energy savings, while keeping the same
communication quality. In the case where the throughputs from every node to every other node are of
the same order, and where the capacities also lie in the same small range, a good choice of those links is
reduced to the problem of finding spanners of the network with low forwarding indices.
In this paper, we consider the case in which the initial graph is a square grid. We consider the
asymptotic case with nlarge. We have two main contributions.
On one side, we know that the forwarding index of the n×ngrid Gnis n3
2, see Proposition 1 [8]. Gn
has 2(n1)22n2edges. An important remark is that the load of the edges is lower in the corner than
in the middle of the grid. Using that, we show that we can build spanners of Gnwith much fewer edges
(only 13/18 72% of the edges) and the same forwarding indices as Gn. We show that the proposed
spanners are close to optimum in the sense that we prove that it is impossible to build spanners with fewer
than 4/3n2edges (66% of the edges).
On the other side, the smallest possible spanner of the n×ngrid Gnis a spanning tree. The for-
warding index of the best spanning tree is asymptotically 3n4
8, see Proposition 2 [8]. When we add edges
and consider spanners with a larger number of edges, the load on the edges decreases, and so does the
forwarding index. In this paper, we study how the forwarding index decreases, when we increase the
number of edges.
The following table summarizes our results:
Spanning tree Spanners Grid
forwarding index 3
8n41
2an4(2 a < n)1
2n31
2n3
lower bound on the number of edges n21'n2+4
9(0.1a)212
9n2
number of edges in the constructions n21n2+4
9a213
9n22n2
Proposition 1 [8] The forwarding index of Gnis asymptotically n3
2.
Proposition 2 [8] For n3, The spanning tree of Gnwith the minimum forwarding index is a tree with
centroid of degree 4 and 4 branches of almost equal sizes. its forwarding index is asymptotically 3n4
8.
RR n° 8643
4Giroire & Pérennes & Tahiri
2 Spanners with the forwarding index of the grid, n3
2, but much
fewer edges
In this section, we first show that a spanner with the forwarding index of the grid has at least 4n2
3=12n2
9
edges. We then provide spanners with 13n2
9edges. But, before, we present some notations that will be
used throughout the paper.
Notations We note by Gn= (Vn, En)the n×nsquare grid, where Vnis the set of Vertices and Enis
the set of edges. A square grid can always be seen as nrows intersecting ncolumns. We name v(r, c)the
vertex at the intersection of row r[n]with column c[n], where [n]denotes the interval of the integer
numbers between 1and n. An edge joining v(r, c)to v(r, c + 1) is named eh(r, c)and an edge joining
v(r, c)to v(r+ 1, c)is named ev(r, c).
Proposition 3 For any Fspanner of Gnsuch that π(F)n3
2,Fmust have, asymptotically, at least 4n2
3
edges.
Proof: Consider Fa spanner of Gnand let Rbe a routing of Fsuch that π(F, R)n3
2. For an integer
l[n], we call load on line l, the sum of the load on the edges ev(l, j )E(F), for j[n]. The load
on line lis 2l(nl)n2as there are ln vertices over line land (nl)nvertices below. If Fhas nxl
edges on line l, there exists at least one of these edges with load at least 2l(nl)n2
nxl. As π(F , R)n3
2, we
should have 2l(nl)n2
nxln3
2.
That is
nxl4l(nl)
n.
Thus, Fshould have at least Pn
l=1
4l(nl)
nvertical edges. The same argument independently holds for
the horizontal edges. Hence, a spanner of the grid, with load lower than n3
2on all edges, has at least
2
n
X
l=1
4l(nl)
n4n2
3edges.
Theorem 1 There exists Fna spanner of Gnsuch that π(Fn)n3
2and its number of edges is asympot-
ically equal to 13n2
9.
Proof: Let us first explain the intuition behind the construction of the spanner of the grid, Fn. We know
from the proof of Proposition 3 the ratio of edges needed in every row or column in order to satisfy the
lower bound. We cut the grid into small squares. Then, according to the position of the square, we use
only the number of needed horizontal edges and vertical edges in each square according to the lower
bound. It turns out that adding only few edges to ensure the connectivity is enough to get a spanner Fn
with a routing Rsuch that π(Fn, R)n3
2.
Construction of Fn.Let kbe an integer number such that 1kn. We divide Gninto small square
grids of size k×k. We do so by partitioning vertices of Gninto (n
k)2sets S(i,j)with i[n
k]and j[n
k]:
S(i,j)={v(r, c)Vn;i1<r
ki, j 1<c
kj}. We call a vertex in S(i,j)that has a neighbour in
Gnoutside S(i,j)a border vertex.
Inria
Grid spanners with low forwarding index 5
Figure 1: Construction of the spanner Fnof Theorem 1, for n= 72, and an example of path of the routing
Rof Fn(from the yellow vertex to the pink vertex).
Let us now describe a spanner Fnthat verifies our theorem. An example of it is shown in Figure 1
in the case of n=k2= 72. Let tbe the function defined on integers by t(x) = d4xk(nxk)k/n2e.
It represents the number of needed columns (respectively rows) for a square that is on the x-th position
horizontally (respectively vertically). We build Fnstarting from a subgraph that has all vertices of Gn
and no edges. For every S(i,j),i, j [n
k], we choose edges to connect vertices in S(i,j)in the following
way:
- we add to Fnall edges ev(r, c)such that (rmod k)∈ {1, . . . , t(i)}(red edges in Figure 1) and
RR n° 8643
6Giroire & Pérennes & Tahiri
- all edges eh(r, c)such that (cmod k)∈ {1, . . . , t(j)}(blue edges in Figure 1);
- then we add to Fnsimple paths just to connect the remaining independent vertices (green edges in
Figure 1).
- We then add all edges that do not have both endpoints in the same set S(i,j)(black edges in 1). We
show in the following that adding all of them is not strictly necessary.
Description of the routing R.We now give a routing of the spanner Fn,R. For every ordered pair
of vertices (v(ra, ca), v(rb, cb)) of Vn, we describe the path connecting v(ra, ca)to v(rb, cb)in R. We
distinguish two types of ordered pairs of vertices:
Type-1 pairs: dra/ke=drb/keor dca/ke=dcb/ke. Notice that this type includes ordered pairs
with vertices that belongs to the same set S(i,j).
Type-2 pairs: All the ordered pairs that do not belong to the first type.
For the Type-1 pairs, Ruses the shortest path routing. For Type-2 pairs, Ruses a three-segment path.
An example of such path is shown in Figure 1. We name ia=dra/ke,ib=drb/ke,ja=dca/keand
jb=dcb/ke:
Step-1: Let im= min(ia, ib, n/k ia, n/k ib)and jm= min(ja, jb, n/k ja, n/k jb). The
first segment is the shortest path from v(ra, ca)to one of the two border vertices of S(ia,ja)that are
on row k(ia1) + t(jm). Among the two vertices, we choose v(rx, cx), which has the smallest
distance to S(ia,jb)(as the first black vertex on the route in Figure 1).
Step-2: Similarly, two border vertices of S(ib,jb)are on column k(ib1) + t(im). Among these
two vertices, v(ry, cy)is the one that has the smallest distance to S(ia,jb)(as the third black vertex
on the route in Figure 1). The second segment will be linking v(rx, cx)to v(ry, cy)by using the
path [v(rx, cx)v(rx, cy)v(rycy)], which is the shortest path from v(rx, cx)to v(rx, cy)composed
of the two direct paths [v(rx, cx)v(rx, cy)], following row rx, and [v(rx, cy)v(rycy)], following
column cy.
Step-3: The third and last segment will be the shortest path from v(ry, cy)to v(rb, cb).
Note that kmay be an arbitrary integer between 1and n. We choose a ksuch that 1kn. For
instance, we may choose k=n1/3.
Number of edges of Fn. Let us compute the number of edges in the spanner, Fn. First the edges used
in the subgraph induced by S(i,j)are all the edges on a row from 1to t(i), all edges on a column from
1to t(j), to which we add the edges that connect the rest of vertices through a spanning tree. Hence the
number of edges in S(i,j)is:
t(i)·k+t(j)·k+ (kt(i))(kt(j))
k2+t(i)t(j)(1)
k2(1 + 16ijk2(nik)(njk)
n4)
k2(1 + 16ijk2
n2+16i2j2k4
n416i2jk3
n316ij2k3
n3)
Inria
Grid spanners with low forwarding index 7
The sum of those edges considering all the subsets S(i,j)(with i[n
k]and j[n
k]) is :
k2
n/k
X
i=1
n/k
X
j=1
(1 + 16ijk2
n2+16i2j2k4
n416i2jk3
n316ij2k3
n3)
k2[n2
k2+16k2
n2(X
i
i)2+16k4
n4(X
i
i2)22·16k3
n3(X
i
i2)(X
i
i)]
k2[n2
k2+16k2
n2·n4
4k4+16k4
n4·n6
9k62·16k3
n3·n5
6k5]
k2[n2
k2+4n2
k2+16n2
9k232n2
6k2]
=13
9n2+o(n2)
The number of the remaining edges is 2n2
k=o(n2), as k1. Therefore, as stated in the theorem,
the number edges of Fnis asymptotically equal to 13n2
9+o(n2).
Load of the edges of Fn.Lets now verify that every edge has an asymptotic load which is not greater
than n3
2+o(n3). Consider an edge eh(r, c)whose incident points are in S(i,j). The number of Type-1
pairs that may use eh(r, c)is bounded by the number of pairs having one endpoint in S(i,ja)and S(i,jb)
for some ja, jb[n
k]and those having one end point in S(ia,j)and S(ib,j)for some ia, ib[n
k]. The
number of these pairs is bounded by 2k2n2=o(n3)(as k2=o(n)).
Then, for Type-2 pairs, we can start by the load induced by the segments of paths described previously
in step-1 and step-3. This load is clearly bounded by the number of pairs having one endpoint inside S(i,j)
and another endpoint outside S(i,j). The number of these pairs is bounded by: 2k2(nk)2=o(n3)(as
k2=o(n)). For Step-2, as the construction of the spanner Fnhas the needed density of edges, the load
is kept below n3
2+o(n3). The same argument holds for the black edges between two adjacent subsets
S(ia,ja)and S(ib,jb). This ends the proof.
3 Spanners with forwarding indices in the range ]n3
2,
3n4
8[and Lower
bounds
We first provide spanners with forwarding indices in the range ]n3
2,3n4
8[in Proposition 4. We then prove
that these spanners have a number of edges of the optimum order, see Proposition 5
3.1 Spanners’ constructions
Proposition 4 Let abe an integer such that, 2an. There exists a spanner Fn(a)of Gnwith
asymptotically n2+4
9a2edges and π(Fn(a)) n4
2a.
Proof: We build a spanner of Gn,Fn(a), in the following way. We divide the grid into a2sectors.
A point is in Sector (i, j)if its coordonnates in the grid (x,y) are such that n
aix < n
a(i+ 1) and
n
ajy < n
a(j+ 1). Each of these sectors has (n/a)2vertices. We call center of the sector (i, j)the
vertex ((i+ 1/2)n/a, (j+ 1/2)n/a). We consider the a×asubgrid linking all the sectors’ centers. We
then connect all the remaining vertices of a sector to its center with a spanning tree. This way, we get
Fn(a). Figure 2 provide a sketch of the construction of the spanner.
RR n° 8643
8Giroire & Pérennes & Tahiri
Figure 2: Spanner of Proposition 4 for n= 21 and a= 3. Edges of the a×agrid are in bold. Edges that
are not in a spanning tree of Gnare in red. Sectors with (n/a)2= 72vertices are separated by dashed
gray lines.
We now build a routing Rfor Fn(a). The demand between two vertices of the same sector are routed
on the tree spanning their sector using the unique shortest path between them. The demand between two
vertices of different sector is first routed to their centers, and then is routed in the a×agrid.
Let us compute the load of the routing R. We first consider the edges of the a×asubgrid. We know
that an a×agrid has a routing with load a3/2(Proposition 1). Thus, we know that it also has a w-routing
of load wa3/2. Each vertex of the a×agrid receives the load of the (n/a)2vertices connected to it.
Thus, we take w= (n/a)2and we obtain a w-routing of the a×agrid of load a3
2(n
a)4=n4
2a.
We then consider an edge that does not belong to the a×agrid. The only paths that can use this
edge are paths going from any vertex of the grid to a vertex of its sector. Thus, its load is smaller that
(n/a)2n2=n3
a2. This load is smaller than the maximum load on the a×agrid as soon as a22awhich
means as soon as a2.
Therefore π(Fn(a), R) = n4
2a.
Let us now consider the number of edges of the spanner Fn(a). The number of edges necessary to
connect all the nodes is n21. If we choose well these edges, we just have to add a2edges to obtain the
a×agrid (see Figure 2, additional edges are in red). Fn(a)thus has n2+a2edges. We can improve the
Inria
Grid spanners with low forwarding index 9
spanner by using the results of Section 2. In Theorem 1, we show that we can find a spanner of an a×a
subgrid with 13
9a2edges and a routing R0with the same load as a full grid with 2a2edges. By doing so,
we get a new spanner Fn(a), with n2+4
9a2edges and π(Fn, R0) = n4
2a.
We can rewrite the result of Proposition 4 to point out the impact of additional edges in general
(Corollary 1) and when we start from a spanning tree (Corollary 2).
Corollary 1 There exist:
- A spanner of Gnwith n2+p2edges, and an asymptotic forwarding index of n4
3p'0.33n4
p;
- A spanner of Gnwith n2+pedges, and an asymptotic forwarding index of n4
3p'0.33 n4
p.
Proof: Direct by Proposition 4 setting p2=4
9a2or p=4
9a2.
Corollary 2 There exists a spanner of forwarding index 1
α
3n4
8, that is a factor αless than the one of the
optimum spanning tree, while using 64
81 α2'0.79α2additional edges compared to a spanning tree.
Proof: Recall that an optimum spanning tree has forwarding index 3n4
8, see Proposition 2. Dividing it
by αmeans getting the forwarding index 3n4
8α=n4
2(4α/3) . This is achieved by the spanner Fn(a), with
a= 4α/3. The spanner has an additional number of edges compared to the spanning tree equal to
4
9(4α/3)2'0.79α2.
3.2 Lower bounds
Proposition 5 There exist no spanners of Gnwith n2+p2edges and a forwarding index less than
1
912
n4
p'0.032n4
p.
Proof: Let us consider a spanner of Gnthat has n2+p2edges. We build a multigraph in the following
way. We start by assigning to every node a weight of 1. Then, while there is still a vertex with degree 1
or 2, we delete this vertex and the edges connecting it to the graph and divide its weight evenly among
its neighbors; in case the removed vertex was of degree 2, we also connect the two neighbors afterwards.
At the end of this process, we get a multigraph Hsuch that the number of its vertices N0and the number
of its edges M0are related by the following equation: N0+p2=M0. Indeed, every time a vertex is
removed in the process leading to H, the number of edges is decreased by 1. Since all the vertices in H
have a degree strictly greater than 2, we have 3
2N0M0. This implies with the previous equation that
3
2N0N0+p2. Hence, we have N02p2. Notice that the total weight is equal to n2. We now apply
the weighted version of the planar separator theorem [6] on H: there exists a partition of the vertices of
Hinto three subsets A,S, and B, such that each of Aand Bhas at most a weight 2n2/3,Shas less
than 6p2p2vertices (The original graph is of a bounded degree 4.) and there are no edges with one
endpoint in Aand another endpoint in B. This directly gives an edge cut of the original graph which has
less than 26p2p2edges and which partitions the original graph’s vertices into two subsets of size at
most 2n2/3. Therefore, any routing of this spanner will induce, at least on one edge of the cut, a load that
is greater than:
1
3n2·2
3n2·1
26p2p2=1
912
n4
p'0.032n4
p.
RR n° 8643
10 Giroire & Pérennes & Tahiri
4 Conclusion
We succeeded at providing spanners of the n×ngrid with a small number of edges for a given forwarding
index. Such spanners are important for energy efficient networks in which the traffic has to be routed in
the network while using a minimum number of equipments. The unused equipments are then turned off
to save energy. We leave as open two problems.
We propose spanners with a number of edges of optimum order for a forwarding index. More pre-
cisely, we have provided spanners of the grid with n2+4
9a2edges and forwarding indices 1
2an4(2 a <
n). We proved that it is impossible to have spanners with the same FI and fewer than 'n2+4
9(0.1a)2
edges. It would be very nice to succeed in filling the gap between the lower bounds and the constructions.
Similarly, we describe spanners with 13/9n2edges and with the same forwarding index of the full
grid Gn. We proved that spanners with such a forwarding index should have at least 12/9n2edges.
Would it be possible to find spanners with such a number of edges?
Last, we focused on a specific network in this work, the square grid. We are also interested by more
general graphs. In particular, the arguments to derive lower bounds can be used for more general planar
graphs with bounded degrees. It would be very interesting to find results and constructions for other
families of planar graphs.
References
[1] Bouabdallah, A. and D. Sotteau, On the edge forwarding index problem for small graphs, Networks
23 (1993), pp. 249–255.
URL http://dx.doi.org/10.1002/net.3230230406
[2] Chabarek, J., J. Sommers, P. Barford, C. Estan, D. Tsiang and S. Wright, Power awareness in net-
work design and routing, in: INFOCOM 2008. The 27th Conference on Computer Communications.
IEEE, IEEE, 2008.
[3] Chiaraviglio, L., M. Mellia and F. Neri, Energy-aware umts core network design, in: The 11th
International Symposium on Wireless Personal Multimedia Communications, 2008.
[4] Chiaraviglio, L., M. Mellia and F. Neri, Reducing power consumption in backbone networks, in:
IEEE International Conference on Communications, 2009, pp. 1–6.
[5] de la Vega, W. F. and Y. Manoussakis, The forwarding index of communication networks with given
connectivity, Discrete Appl. Math. 37-38 (1992), pp. 147–155.
URL http://dx.doi.org/10.1016/0166-218X(92)90129-X
[6] Djidjev, H. N., On the problem of partitioning planar graphs, SIAM Journal on Algebraic Discrete
Methods 3(1982), pp. 229–240.
[7] Fernandez de la Vega, W. and L. M. Gordones, The forwarding indices of random graphs, Random
Structures and Algorithms 3(1992), pp. 107–116.
URL http://dx.doi.org/10.1002/rsa.3240030108
[8] Giroire, F., D. Mazauric and J. Moulierac, “Energy Efficient Routing by Switching-Off Network
Interfaces,” IGI Global, 2012 pp. 207–236.
[9] Giroire, F., D. Mazauric, J. Moulierac and B. Onfroy, Minimizing routing energy consumption: from
theoretical to practical results, in: IEEE/ACM International Conference on Green Computing and
Communications (GreenCom’10), Hangzhou, China, 2010.
Inria
Grid spanners with low forwarding index 11
[10] Giroire, F., J. Moulierac, T. Phan and F. Roudaut, Minimization of network power consumption with
redundancy elimination, in: IFIP Networking (2012), pp. 247–258.
[11] Giroire, F., J. Moulierac and T. K. Phan, Optimizing rule placement in software-defined networks for
energy-aware routing, in: Proceedings of the IEEE Global Communications Conference (GLOBE-
COM), Austin, United States, 2014.
[12] R, J. Baliga, R. Ayre, K. Hinton and W, Energy consumption in IP networks, in: Optical Communi-
cation, 2008. ECOC 2008. 34th European Conference on, 2008.
[13] Restrepo, J. C., C. G. Gruber and C. M. Machuca, Energy profile aware routing, in: Communications
Workshops, 2009. ICC Workshops 2009. IEEE International Conference on, IEEE, 2009, pp. 1–5.
[14] Xu, J.-M. and M. Xu, The forwarding indices of graphs – a survey, CoRR abs/1204.2604 (2012).
RR n° 8643
RESEARCH CENTRE
SOPHIA ANTIPOLIS – MÉDITERRANÉE
2004 route des Lucioles - BP 93
06902 Sophia Antipolis Cedex
Publisher
Inria
Domaine de Voluceau - Rocquencourt
BP 105 - 78153 Le Chesnay Cedex
inria.fr
ISSN 0249-6399
... We now study graphs with a constraint on the degree (∆ will denote the maximum degree). The motivation comes from telecommunication & real interconnection networks for which the node degree is often small, see for example [19,8]. ...
Article
Full-text available
The (edge) forwarding index of a graph is the minimum, over all possible rout-ings of all the demands, of the maximum load of an edge. This metric is of a great interest since it captures the notion of global congestion in a precise way: the lesser the forwarding-index, the lesser the congestion. In this paper, we study the following design question: Given a number e of edges and a number n of vertices, what is the least congested graph that we can construct? and what forwarding-index can we achieve? Our problem has some distant similarities with the well-known (∆, D) problem, and we sometimes build upon results obtained on it. The goal of this paper is to study how to build graphs with low forwarding indices and to understand how the number of edges impacts the forwarding index. We answer here these questions for different families of graphs: general graphs, graphs with bounded degree, sparse graphs with a small number of edges by providing constructions, most of them asymptotically optimal. For instance, we provide an asymptotically optimal construction for (n, n + k) cubic graphs-its forwarding index is ∼ n 2 3k log 2 (k). Our results allow to understand how the forwarding-index drops when edges are added to a graph and also to determine what is the best (i.e least congested) structure with e edges. Doing so, we partially answer the practical problem that initially motivated our work: If an operator wants to power only e links of its network, in order to reduce the energy consumption (or wiring cost) of its networks, what should be those links and what performance can be expected?
Conference Paper
The (edge) forwarding index of a graph is the minimum, over all possible routings of all the demands, of the maximum load of an edge. This metric is of a great interest since it captures the notion of global congestion in a precise way: the lesser the forwarding-index, the lesser the congestion. In this paper, we study the following design question: Given a number e of edges and a number n of vertices, what is the least congested graph that we can construct? and what forwarding-index can we achieve? Our problem has some distant similarities with the well-known \((\varDelta ,D)\) problem, and we sometimes build upon results obtained on it. The goal of this paper is to study how to build graphs with low forwarding indices and to understand how the number of edges impacts the forwarding index. We answer here these questions for different families of graphs: general graphs, graphs with bounded degree, sparse graphs with a small number of edges by providing constructions, most of them asymptotically optimal. For instance, we provide an asymptotically optimal construction for \((n,n+k)\) cubic graphs - its forwarding index is \(\sim \frac{n^2}{3k} \log _2(k)\). Our results allow to understand how the forwarding-index drops when edges are added to a graph and also to determine what is the best (i.e. least congested) structure with e edges. Doing so, we partially answer the practical problem that initially motivated our work: If an operator wants to power only e links of its network, in order to reduce the energy consumption (or wiring cost) of its networks, what should be those links and what performance can be expected?
Article
Full-text available
Software-defined Networks (SDN), in particular OpenFlow, is a new networking paradigm enabling innovation through network programmability. Over past few years, many applications have been built using SDN such as server load balancing, virtual-machine migration, traffic engineering and access control. In this paper, we focus on using SDN as an approach for energy-aware routing (EAR). Since traffic load has a small influence on power consumption of routers, EAR allows to put unused links into sleep mode to save energy. SDN can collect traffic matrix and then computes routing solutions satisfying QoS while being minimal in energy consumption. However, prior works on EAR have assumed that the table of OpenFlow switch can hold an infinite number of rules. In practice, this assumption does not hold since the flow table is implemented in Ternary Content Addressable Memory (TCAM) which is expensive and power-hungry. In this paper, we propose an optimization method to minimize energy consumption for a backbone network while respecting capacity constraints on links and rule space constraints on routers. In details, we present an exact formulation using Integer Linear Program (ILP) and introduce efficient greedy heuristic algorithm for large networks. Based on simulations, we show that using this smart rule space allocation, it is possible to save almost as much power consumption as in the classical EAR approach.
Article
Full-text available
Several studies exhibit that the traffic load of the routers only has a small influence on their energy consumption. Hence, the power consumption in networks is strongly related to the number of active network elements, such as interfaces, line cards, base chassis,... The goal thus is to find a routing that minimizes the (weighted) number of active network elements used when routing. In this paper, we consider a simplified architecture where a connection between two routers is represented as a link joining two network interfaces. When a connection is not used, both network interfaces can be turned off. Therefore, in order to reduce power consumption, the goal is to find the routing that minimizes the number of used links while satisfying all the demands. We first define formally the problem and we model it as an integer linear program. Then, we prove that this problem is not in APX, that is there is no polynomial-time constant-factor approximation algorithm. We propose a heuristic algorithm for this problem and we also prove some negative results about basic greedy and probabilistic algorithms. Thus we present a study on specific topologies, such as trees, grids and complete graphs, that provide bounds and results useful for real topologies. We then exhibit the gain in terms of number of network interfaces (leading to a global reduction of approximately 33 MWh for a medium-sized backbone network) for a set of existing network topologies: we see that for almost all topologies more than one third of the network interfaces can be spared for usual ranges of operation. Finally, we discuss the impact of energy efficient routing on the stretch factor and on fault tolerance.
Conference Paper
Full-text available
According to several studies, telecom operators account for a not negligible power consumption, and several initiatives are being put into place to reduce the power consumption of the ICT sector in general. To this goal, we propose a novel ap- proach to switch off some portions of the UMTS core network while still guaranteeing full connectivity and maximum link utilization. After showing that the problem falls in the class of capacitated multi-commodity flow problems, and therefore it is NP-complete, we propose some heuristic algorithms to solve it. Results show that it is possible to reduce the number of links and nodes currently used by up to 30% and 50% respectively during off-peak hours while offering the same service quality. I I NTRODUCTION Power consumption in general, and of ITC technologies in particular, has become a key issue during the last few years. The ratio of power demand versus power resources is con- stantly growing, and energy costs are increasing at a constant rate. The electricity cost jumped up of about 35% in Italy dur- ing the period 2004-2007 (4). Moreover, Green House Gases (GHG) emissions have a negative impact on the world climate change (8), and people are becoming more conscious about problems that will arise in the near future due to this. According to a number of studies, ICT alone is responsible for a percentage which varies from 2% to 10% of the world power consumption (7), due to the ever increasing diffusion of electronic devices. In this scenario, the power consumption of telecommunication networks is not negligible. For example, Telecom Italia consumes more than 2 TWh of energy, repre- senting the second national user (6) after the national railway operator. More recently, telecom operators have begun to adopt equip- ments that are able to be enter into standby mode (1). In this paper, we consider the packet switching domain of a typical UMTS architecture. Given the network topology and a traffic demand from the Radio Access Network, we evaluate the possibility to turn off some elements (nodes and links) un- der connectivity and Quality of Service (QoS) constraints. The goal is to minimize the total power consumption of the core network, in which usually resource over-provisioning is large. We investigate some simple optimization algorithms. In partic- ular, we selectively power off nodes and links of the topology following different strategies. Results show that it is possible to reduce the percentage of nodes and links actually powered on up to 30% and 50% respectively, while guaranteeing that the resource utilization is still below a given threshold. e.g., 50%. The paper is organized as follows: Sect. II presents the sce- nario and the problem formulation; Sect. III describes the im- plemented heuristics, and results are presented in Sect. IV. Fi- nally, conclusions are drawn in Sect. V.
Conference Paper
Full-text available
According to several studies, the power consumption of the Internet accounts for up to 10% of the worldwide energy consumption, and several initiatives are being put into place to reduce the power consumption of the ICT sector in general. To this goal, we propose a novel approach to switch off network nodes and links while still guaranteeing full connectivity and maximum link utilization. After showing that the problem falls in the class of capacitated multi-commodity flow problems, and therefore it is NP-complete, we propose some heuristic algorithms to solve it. Simulation results in a realistic scenario show that it is possible to reduce the number of links and nodes currently used by up to 30% and 50% respectively during off-peak hours, while offering the same service quality.
Conference Paper
Recently, energy-aware routing has gained increasing popularity in the networking research community. The idea is that traffic demands are aggregated over a subset of the network links, allowing other links to be turned off to save energy. In this paper, we propose GreenRE - a new energy-aware routing model with the support of the new technique of data redundancy elimination (RE) . This technique, enabled within the routers, can identify and remove repeated content from network transfers. Hence, capacity of network links are virtually increased and more traffic demands can be aggregated. Based on our real experiments on Orange Labs platform, we show that performing RE consumes some energy. Thus, while preserving connectivity and QoS, it is important to identify at which routers to enable RE and which links to turn off so that the power consumption of the network is minimized. We model the problem as an Integer Linear Program and propose a greedy heuristic algorithm. Simulations on several network topologies show that GreenRE can gain further 30% of energy savings in comparison with the traditional energy-aware routing model.
Article
The results in this paper are closely related to the effective use of the divide-and-conquer strategy for solving problems on planar graphs. It is shown that every planar graph can be partitioned into two or more components of roughly equal size by deleting only $O ( \sqrt{n} )$ vertices, and such a partitioning can be found in $O ( n )$ time. Some of the theorems proved in the paper are improvements on the previously known theorems while others are of more general form. An upper bound for the minimum size of the partitioning set is found.
Article
A routing R of a graph G is a set of n(n − 1) elementary paths R(u, v) specified for all ordered pairs (u, v) of vertices of G. The vertex-forwarding index ξ(G) of G, is defined by Where ξ(G, R) is the maximum number of paths of the routing R passing through any vertex of G and the minimum is taken over all the routings of G. Let Gp denote the random graph on n vertices with edge probability p and let m = np. It is proved among other things that, under natural growth conditions on the function p = p(n), the ratio Tends to 1 in probability as n tends to infinity.
Article
For a given graph G of order n, a routing R is a set of n(n − 1) elementary paths specified for every ordered pair of vertices in G. The edge forwarding index of a network (G,R), denoted π(G,R) is the maximum number of paths of R going through any edge e of G. The edge forwarding index of G, denoted π(G), is the minimum of π(G,R) taken over all the possible routings R of G. Given n ≤ 15 and Δ ≤ n − 1 we determine πΔ,n, the minimum of π(G) taken over all graphs G of order n with maximum degree at most Δ. This is known as the edge forwarding index problem. © 1993 by John Wiley & Sons, Inc.
Conference Paper
Energy efficient communication devices are essential to minimize the operational cost of future networks and to reduce the negative effects of global warming. In this paper we propose a novel energy reduction approach on network level that takes load-dependent energy consumption information of communication equipment into account. Case study calculation results show that energy savings of more than 35% and with it operational cost can be saved by applying energy profile aware routing.