Protection Schemes for Two Link Failures

in Optical Networks
Salah A. Aly and Ahmed E. Kamal
Department of Electrical and Computer Engineering
Iowa State University, Ames, IA 50011, USA
Email: {salah,kamal}@iastate.edu
November 12, 2008
arXiv:0811.1693v1 [cs.IT] 11 Nov 2008

Abstract—In this paper we develop network protection to maximize the throughput [1], [10]. Also, it is robust against
schemes against two link failures in optical networks. The packet losses and network failures [5], [11], [6].
motivation behind this work is the fact that the majority of Network protection against single and multiple link failures
all available links in an optical network suffer from single and
double link failures. In the proposed network protection schemes, using adding extra protection paths has been introduced in [7],
NPS2-I and NPS2-II, we deploy network coding and reduced [8]. The source nodes are able to combine their data into
capacity on the working paths to provide backup protection extra paths (backup protection paths) that are used to protect
paths. In addition, we demonstrate the encoding and decoding all signals on the working paths plain carrying data from all
aspects of the proposed schemes. sources. In both cases, p-cycles has been used for protection
Index Terms—Network Protection, Optical Networks.
against single and multiple link failures.
In this paper we design two schemes for network protection
I. I NTRODUCTION
against one and two failed links in a network with n disjoint
One of the main services of operation networks that must be working paths: NPS2-I and NPS2-II. The approach is based
deployed efficiently is reliability. In order to deploy a reliable on network coding data originated by the sources. We assume
networking strategy, one needs to protect the transmitted sig- that network capacity will be reduced by a partial factor in
nals over unreliable links. Link failures are common problems order to achieve the required protection. Several advantages
that might occur frequently in single and multiple operating of NPS2-I and NPS2-II can be stated as:
communication circuits. In network survivability and network • The data sent from the sources are protected without
resilience, one needs to design efficient strategies to overcome adding extra paths. Two paths out of the n disjoint
this dilemma. Optical network survivability techniques are working paths will carry encoded data, and hence they
classified as pre-designed protection and dynamic restora- are protection paths.
tion [13], [8]. The approach of using pre-designed protection • The encoding and decoding operations are achieved with
aims to reserve enough bandwidth such that when a failure less computational cost at both the sources and receivers.
occurs, backup paths are used to reroute the transmission The recovery from failures is achieved immediately with-
and be able to recover the data. Examples of this type are out asking the senders to retransmit the lost data.
1-1 and 1-N protections [2], [9]. In dynamic restoration • The normalized network capacity is (n − 2)/n, which
reactive strategies, capacity is not reserved. However, when is near-optimal in case of using a large number of
the failure occurs, dynamic recovery is used to recover the connections.
data transmitted in the links that are suffered from failures.
This technique does not need preserved resources or provision II. N ETWORK M ODEL
of extra paths that work in case of failure. In this work we In this section we present the network model and basic
will provide several strategies of dynamic restoration based on terminology.
coding and reduced distributed fairness capacities. i) Let N be a network represented by an abstract graph
Network coding is a powerful tool that has been recently G = (V, E), where V is a set of nodes and E be a set
used to increase the throughput, capacity, and performance of undirected edges. Let S and R be sets of independent
of communication networks. Information theoretic aspects of sources and destinations, respectively. The set V = V ∪
network coding have been investigated in [12], [1]. Network S ∪ R contains the relay, source, and destination nodes.
coding allows the intermediate nodes not only to forward Assume for simplicity that |S| = |R| = n, hence the set
packets using network scheduling algorithms, but also en- of sources is equal to the set of receivers.
code/decode them using algebraic primitive operations, see [1], ii) A path (connection) is a set of edges connected together
[4], [12] and references therein. As an application of network with a starting node (sender) and an ending node (re-
coding, data loss because of failures in communication links ceiver).
can be detected and recovered if the sources are allowed to
perform network coding operations. Network coding is used Li = {(si , e1i ), (e1i , e2i ), . . . , (e(m)i , ri )},
 2

where 1 ≤ i ≤ n and (e(j−1)i , eji ) ∈ E for some integer x1

s1 r1
m.
iii) The node can be a router, switch, or an end terminal x2
depending on the network model N and the transmission s2 r2
layer.
iv) L is a set of paths L = {L1 , L2 , . . . , Ln } carrying the data x i | x 1 +x 2 +…+x i-1 +x i+1 +…+x n
from the sources to the receivers as shown in Fig. 2. All si ri
connections have the same bandwidth, otherwise a con-
nection with high bandwidth can be divided into multiple x j | x 1 +x 2 +…+x j-1+x j+1 +…+x n
connections, each of which has a unit capacity. There are sj rj
exactly n connections. For simplicity, we assume that the
number of sources is less than or equal to the number of s3 Xn-1
r3
available paths. A sender with a high capacity can divide
its capacity into multiple unit capacities, each of which xn
sn rn
has its own path. The failure on a link Li may occur due
to the network circumstance such as a link replacement, Fig. 1. Network protection against link/path failures using reduced capacity
overhead, etc. We do not address in this work the failure and network coding. Two paths out of n disjoint working paths carry encoded
cause. However, we assume that there are one or two data for protection against two link failures.
failures in the working paths and the protection strategy
is able to protect/recover it.
v) Each sender si ∈ S will transmit its own data xi to a Definition 2: The working paths of a network with n
receiver ri through
P a connection Li . Also, si will transmit connection paths are the paths that carry unencoded data traffic
encoded data ni xi to ri at a different time slot if it is under normal operations. The Protection paths are paths that
assigned to send the encoded data. provide alternate backup paths to carry encoded data traffic in
vi) The data from all sources are sent in sessions. Each case of failures. A protection scheme ensures that data sent
session has a number of time slots n. Hence tℓδ is a value from the sources will reach the receivers in case of failure
at round time slot ℓ in session δ. incidences on the working paths.
vii) In this model N if we consider only a single link failure, Every sender si prepares a packet packetsi →ri to send to
then it is sufficient to apply the encoding and decoding the receiver ri . The packet contains the sender’s ID, data xℓi ,
operations over a finite field with two elements, F2 = and a round time for every session tℓδ for some integers δ and
{0, 1}. However, if there are double failures, then a finite ℓ. We have two types of packets:
field with higher alphabets is required. i) Packets sent without coding, in which the sender does not
viii) There are at least two receivers and two senders with at need to perform any coding operations. For example, in
least two disjoint paths. Otherwise, the protection strategy case of packets sent without coding, the sender si sends
can not be deployed for a single path, which it can not the following packet to the receiver ri .
protect itself.
We can define the network capacity in the light of min-cut packetsi →ri := (IDsi , xℓi , tℓδ ) (3)
max-flow information theoretic view [1]. ii) Packets sent with encoded data, in which the sender needs
Definition 1: The capacity of a connecting path Li between to send other sender’s data. In this case, the sender si
si and ri is defined by sends the following packet to receiver ri :
n

1, Li is active;
ci =
X
(1) packetsi →ri := (IDsi , xℓj , tℓδ ). (4)
0, otherwise.
j=1,j6=i
The total capacity is given by the summation of all paths’
The value yiℓ = nj=1,j6=i xℓj is computed by every sender
P
capacities. What we mean by an active link is that the receiver
is able to receive and process signals/messages throughout this si , by collecting the data from all other senders and
link. encoding them using the bit-wise operation.
In either case, the sender has a full capacity in the connection
Clearly, if all links are active then the total capacity is n and path Li .
normalized capacity is 1. In general the normalized capacity
The protection path that carries the encoded data from
of the network for the active and failed links is computed as:
all sources is used for the data recovery in case of failure.
n
1X Assuming the encoding operations occur in the same round
CN = ci . (2) time of a particular session, every source si adds its value, for
n i=1
1 ≤ i ≤ n. Therefore, the encoded data over the protection
path is yi = nj=1,i6=j xj . The decoding operations are done
P
We define the working paths and protection paths as fol-
lows: at every receiver ri by adding the data xi received over the
 3

working path Li . The node rk with failed connection Lk will In the case of two failures, the receivers will be able to solve
be able to recover the data xk . Assuming all operations are two linearly independent equations in two unknown variables.
achieved over the binary finite field F2 . Hence we have For instance, assume the two failures occur in paths number
n
X two and four. Then the receivers will be able to construct two
xk = yi − xj . (5) equations with cofficients
j=1,i6=j
 
1 1
(10)
III. P ROTECTIONS U SING D EDICATED PATHS (NPS2-I) α α3
In this section we develop a network protection scheme Therefore, we have
(NPS2-I) for two link failures in optical networks. The protec-
xℓ2 + xℓ4 (11)
tion scheme is achieved using network coding and dedicated
paths. Assume we have n connections carrying data from a set αxℓ2 + α3 xℓ4 (12)
of n sources to a set of n receivers. All connections represent One can multiply the first equation by α and subtract the two
disjoint paths, and the sources are independent of each other. equations to obtain value of xℓ4 .
The authors in [7], [3] introduced a model for optical network We notice that the encoded data symbols yjℓ and ykℓ are fixed
protection against a single link failure using an extra and per one session transmission but it is varied for other sessions.
dedicated paths provisioning. In this model NPS2-I we extend This means that the path Lj is dedicated to send all encoded
this approach to two link failures. data yj1 , yj2 , . . . , yjn .
We will provide two backup paths to protect against any Lemma 3: The normalized capacity of NPS2-I of the net-
two disjoint links, which might experience failures. These two work model N described in (6) is given by
protection paths can be chosen using network provisioning.
The protection paths are fixed for all rounds per session, C = (n − 2)/n. (13)
but they may vary among sessions. For example, sender si Proof: There are n rounds in every session. Also, we have
transmits a message xℓi to a receiver ri at time tℓδ in round n connections per a round time. There exist two connections
time ℓ in session δ. This process is explained in Scheme (6) which carry backup data for protection, hence there are n − 2
as: connections that carry working data. Therefore, the normalized
capacity is given as:
round time session 1 ...
C = (n − 2)n/n2 ,
1 2 3 ... n ...
s1 → r1 x11 x1 2
x13
... xn1 ... which gives the result.
s2 → r2 x12 x22 x32 ... xn2 ... In NPS2-I there are three different scenarios for two link
s3 → r3 x13 x23 x33 ... xn3 ... failures, which can be described as follows:
.. .. .. .. .. i) If the two link failures occur in the backup protection
. . . . . ... paths Lj and Lk , then no recovery operations are required
(6)
si → ri x1i x2i x3i . . . xni ... at the receivers side. The reason is that these two paths
sj → rj yj1 yj2 yj3 . . . yjn ... are used for protections, and all other working paths will
sk → rk yk1 yk2 yk3 . . . ykn ... convey the data from the senders to receivers.
si+1 → ri+1 x1i+1 x2i+1 x3i+1 . . . xni+1 ... ii) If the two link failures occur in one backup protection
.. .. .. .. .. .. path say Lj and one working path Li , then recovery
. . . . . . ...
sn → rn x1n x2n x3n . . . xnn ... operations are required. The receiver ri must recover its
data using one of the protection paths.
All yjℓ ’s are defined as: iii) If the two link failures occur in two working paths, then
n n
in this case the two protection paths are used to recover
yjℓ =
X
aℓi xℓi and ykℓ =
X
bℓi xℓi . (7) the lost data. The idea of recovery in this case is to build
i=1,i6=j6=k i=1,i6=k6=j
a system of two equations with two unknown variables.

The coefficients aℓi and bℓi are chosen over a finite field Fq IV. P ROTECTION AGAINST T WO L INK FAILURES
with q > n − 2, see [3] for more details. One way to choose (NPS2-II)
these coefficients is by using the follow two vectors. In this section we will provide an approach for network pro-
  tection against two link failures in optical networks. We deploy
1 1 1 ... 1
(8) network coding and distribute capacity over the working paths.
1 α α2 . . . αn−3 We will compute the network capacity in this approach. In [3]
Therefore, the coded data is we will illustrate the tradeoff and implementation aspects of
n
X n
X this approach, where there is enough space for details.
mod n−2 ℓ We assume that there is a set of n connections that need
yjℓ = xℓi and ykℓ = αi xi . (9)
i=1,i6=j6=k i=1,i6=k6=j to be protected with %100 guarantee against single and two
 4

link failures. Assume Fq is a finite field with q elements. For s1 r1

simplicity we consider n is an even number.
s2 r2
A. Two Link Failures, Achieving (n − 2)/n Capacity
Data Data
Let xℓi be the data sent from the source si at round time ℓ distributor s3 r3 distributor
in a session at time tℓδ . Also, assume the encoded data yi =
P n ℓ
j=1,j6=i xj . Put differently:
.. …..

yi = xℓ1 ⊕ xℓ2 ⊕ . . . ⊕ xℓi6=j ⊕ . . . ⊕ xℓn . (14)

sn rn
The protection scheme NPS2-II runs in sessions as explained
below. Every session has at most one single failure through out Fig. 2. The encoding and decoding operations are done at the data distributor
its each round time. As shown in Scheme (15), the protection and collector, respectively.
matrix for the first session is given by the following protection
code:
be an index of an arbitrary session that has two link failures.
round time session 1 We have n paths that have capacity n. Also, we have ℓ = n/2
1 2 3 ... ... ℓ round times, in which each round time has n − 2 capacity in
the working paths. Therefore the total capacity is given by
s1 → r1 y11 x11 x21 ... ... xℓ−1
1
s2 → r2 y21 x12 x22 ... ... xℓ−1
2
(n − 2)(ℓ) = (n2 − 2n)/2. (18)
s3 → r3 x13 y32 x23 ... ... xℓ−1
3 By normalizing this value with the total capacity nℓ, then the
s4 → r4 x14 y42 x24 ... ... xℓ−1
3 result follows.
.. .. .. .. .. .. .. (15)
. . . . . . . The network protection strategy NPS2-II against one or
si → ri xi 1
xi 2
... yi j ℓ−1
. . . xi two link failures is deployed in two processes: Encoding and
1
si+1 → ri+1 xi+1 xi+1 2 j
. . . yi+1 . . . xi+1 ℓ−1 decoding operations. The encoding operations are performed
.. .. .. .. .. .. .. at the set of sources, in which one or two sources will send
. . . . . . . the encoded data depending on the number of failures. The
sn−1 → rn−1 x1n−1 x2n−1 x3n−1 . . . . . . yn−1 ℓ
decoding operations are performed at the receivers’ side, in
1 2 3
sn → rn xn xn xn ... ... ynℓ which receivers with failed links had to receive all other
where receivers’ data in order to recover their own data. Depending
on NPS2-I or NPS2-II the receivers will experience some delay
2(ℓ−1) n
X X before they can actually decode the packets. The transmission
ykℓ = aℓ−1
i xℓ−1
i + aℓi xℓi is done in rounds, hence linear combinations of data has to
i=1 i=2ℓ+1
be from the same round time. This can be achieved using the
for (2ℓ − 1) ≤ k ≤ 2ℓ, 1 ≤ ℓ ≤ n/2. (16) round time that is included in each packet sent by a sender.
All coefficients are taken from Fq for q > n − 2, see [3] for Assume there are data collectors S and R at the senders
more details. Also, the two vectors shown in 8 can be used in and receivers, respectively. They can be a sender (receiver)
this case. We note that the data symbols in NPS2-II are sent node to send (receiver) encoded data, see [3].
in independent sessions. This means that every session has its
own recovery scheme. Also, two failures occur in only two Encoding Process: The encoding operations are for each
connections during the session round times. Hence the sender round per a session.
si sends the message xji for all 1 ≤ j ≤ ℓ − 1 and 1 ≤ i ≤ n • The source nodes send a copy of their data to the data
during the first session. One can always change the round time distributor S, then S will decide which source will send
of the encoded data ykℓ and the data xji for any round time j the encoding data ykℓ and all other sources will send their
in the same session. own data xℓi . This process will happen in every round
Now, we shall compute the normalized capacity of NPS2-II during transmission time.
for the network N at one particular session; the first session. • The encoding is done by linear operation of sources’
The capacity is calculated using the well-known min-cut max- coefficients which is the fastest arithmetic operation that
flow theorem [1]. can be performed among all sources’ data.
Theorem 4: The optimal fairness distributed normalized • The server S will change the sender si that should send
capacity of NPS2-II shown in Scheme (15) is given by the encoded data yiℓ in every round of a given session for
the purpose of fairness and distributed capacities.
C = (n − 2)/n. (17)
Proof: Let n be the number of sources, each of which Decoding Process: The objective of the decoding and recovery
has a unit capacity in the connection Li from si to ri . Let j process is to withhold rerouting the signals or the transmitted
 5

packets due to link failures, see [2], [3], [8]. However, we [6] S. Jaggi, M. Langberg, S. Katti, T. Ho, D. Katabi, and M. Medard.
provide strategies that utilize network coding and reduced Resilient network coding in the presence of byzantine adversaries. In
Proc. IEEE INFOCOM, 2007.
capacity at the source nodes. We assume that the receiver [7] A. E. Kamal. 1+N protection in optical mesh networks using network
nodes are able to perform decoding operations using a data coding on p-cycles. In Proc. of the IEEE Globecom, 2006.
collector R. [8] A. E. Kamal. 1+N network protection for mesh networks: network
coding-based protection using p-cycles. submitted to IEEE Journal of
We assume there is a data distributor R that will collect Communication, 2008.
data from all working and protection paths and is able to [9] A. E. Kamal. A generalized strategy for 1+N protection. In Proc. of
perform the decoding operations. In this case we assume that the IEEE International Conference on Communications (ICC), 2008.
[10] R. Koetter and M. Medard. An algebraic approach to network coding.
all receivers R have available shared paths with the data IEEE/ACM transactions on networking, 2003.
collector R. At the receivers side, if there are two failures [11] D. S. Lun, N. Ranakar, R. Koetter, M. Medard, E. Ahmed, and H. Lee.
in paths Lj and Lk , then there are several situations. Achieving minimum-cost multicast: A decentralized approach based on
network coding. In In Proc. the 24th IEEE INFOCOM, volume 3, pages
• If the paths Lj and Lk carry unencoded data (working 1607– 1617, March 2005.
paths), then the data distributor R must query all other [12] E. Soljanin and C. Fragouli. Network codinginformation flow perspec-
tive. 2007.
nodes in order to recover the failed data. In this case rk [13] D. Zhou and S. Subramaniam. Survivability in optical networks. IEEE
and rj must query R to retrieve their lost data. network, 14:16–23, Nov./Dec. 2000.
• If the path Lk carries encoded data yk (protection path)
and Lj carries unencoded data (working path), then data
collector R must query all other receivers in order to
perform decoding, and rj receives the lost data xℓj .
• If the paths Lj and Lk carry encoded data (they are both
protection paths), then no action is required.

V. C ONCLUSION
In this paper we presented network protection schemes
NPS2-I and NPS2-II against single and double link failures in
optical networks. We showed that protecting two failures can
be achieved using network coding and reduced capacity. The
normalized capacity of the proposed schemes is (n − 2)/n,
which is near optimal for a large number of connections.
Extended version of this paper will appear in [3], where
protection against t multiple failures is investigated.

VI. ACKNOWLEDGMENTS
This research was supported in part by grants CNS-0626741
and CNS-0721453 from the National Science Foundation, and
a gift from Cisco Systems. S A. A would like to thank Prof.
A E. K for his support and guidance. He is also grateful to
Prof. E. Soljanin and Prof. A. Klappenecker for their kindness
and for training him how to inscribe high quality research
papers.

R EFERENCES
[1] R. Ahlswede, N. Cai, S.-Y. R. Li, and R. W. Yeung. Network information
flow. IEEE Trans. Inform. Theory, 46:1204–1216, 2000.
[2] S. A. Aly and A. E. Kamal. Network protection codes against link
failures using network coding. In Proc. IEEE GlobelComm ’08, New
Orleans, LA, December 2008.
[3] S. A. Aly and A. E. Kamal. On the construction of network protection
codes against network failures. IEEE Journal on Selected Areas in
Communciation (JSAC) , on submission, 2008.
[4] C. Fragouli, J. Le Boudec, and J. Widmer. Network coding: An instant
primer. ACM SIGCOMM Computer Communication Review, 36(1):63–
68, 2006.
[5] T. Ho, R. Koetter, M. Medard, D. Karger, and M. Effros. The benefits
of coding over routing in a randomized setting. In Proc. of the IEEE
International Symposium on Information Theory (ISIT03), page 442,
Yokohama, Japan, June 2003.