02.12.2012 Views

j - AGH University of Science and Technology

j - AGH University of Science and Technology

j - AGH University of Science and Technology

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

12th GI/ITG CONFERENCE ON MEASURING, MODELLING AND<br />

EVALUATION OF COMPUTER AND COMMUNICATION SYSTEMS<br />

3rd POLISH-GERMAN TELETRAFFIC SYMPOSIUM<br />

AVAILABILITY ASSESSMENT OF RESILIENT NETWORKS<br />

Piotr Chołda*, Andrzej Jajszczyk**<br />

Department <strong>of</strong> Telecommunications, <strong>AGH</strong> <strong>University</strong> <strong>of</strong> <strong>Science</strong> <strong>and</strong> <strong>Technology</strong>,<br />

Al. Mickiewicza 30, 30-059 Kraków, Pol<strong>and</strong><br />

*E-mail: cholda@kt.agh.edu.pl<br />

**E-mail: jajszczyk@kt.agh.edu.pl<br />

Abstract<br />

In the paper we present approaches to calculate single connections availability. Approximate<br />

calculations for the most representative protection topologies as well as for a particular case <strong>of</strong><br />

restoration in MultiProtocol Label Switching (MPLS) are presented <strong>and</strong> discussed. On this basis some<br />

simple conclusions concerning the way in which paths should be allocated are drawn.<br />

Keywords<br />

Availability, protection, restoration<br />

1. INTRODUCTION<br />

The problem <strong>of</strong> network reliability assurance has been present since the first<br />

network was built. We propose some methods for the calculation <strong>of</strong> one <strong>of</strong> the<br />

basic <strong>and</strong> most important reliability measures used by network engineers <strong>and</strong><br />

planners, the availability. It can be used for many different purposes: in general<br />

network performance evaluation as well as an important factor during the path<br />

computation phase. In the latter case it could be incorporated in dynamic protocols,<br />

providing a new important dimension in path computation algorithms. Although it<br />

is relatively easy to calculate the availability for simple ring networks, exact<br />

analytical calculations in the case <strong>of</strong> mesh networks lead to serious difficulties: the<br />

changing environment caused by routing should be taken into account <strong>and</strong> the fact<br />

that the capacity is shared among different connections. These factors hinder the<br />

enumeration process <strong>of</strong> probabilistic events related to failures <strong>and</strong> recovery<br />

procedures. Therefore, the availability <strong>of</strong> selected paths <strong>and</strong> connections, <strong>and</strong> not <strong>of</strong><br />

the whole network, is usually considered [1].


Piotr Chołda, Andrzej Jajszczyk<br />

To show how reliability measures are calculated <strong>and</strong> how to compare different<br />

recovery procedures, we chose only a few typical protection schemes. We were<br />

also paying attention to simplicity <strong>and</strong> transparency <strong>of</strong> the relevant formulas. We<br />

can note that there are also other reliability measures, e.g., the reliability<br />

polynomial for the whole network, down time ratio or average traffic loss.<br />

However, the availability is the most common measure <strong>and</strong> can be used both in<br />

very simple as well as very sophisticated (e.g., the most recent [2]) analyses.<br />

It is very important to note that many different types <strong>of</strong> availability are<br />

defined. Formally, the so called instantaneous availability is defined in terms <strong>of</strong> the<br />

ITU-T Recommendation [3] as the probability that an item is in an up state at a<br />

given instant <strong>of</strong> time. However, for our calculations we will use the concept <strong>of</strong><br />

asymptotic (steady-state) availability, which means the limit, if such exists, <strong>of</strong> the<br />

instantaneous availability when the time tends to infinity. During calculations, it is<br />

assumed that the failures <strong>of</strong> the elements are statistically-independent. The<br />

availability defined in such a way could be understood as an average ratio <strong>of</strong> time<br />

in which an item or a connection works properly, to the whole time.<br />

2. AVAILABILITY OF THE CONNECTION WITHOUT<br />

PROTECTION<br />

An unprotected connection consists only <strong>of</strong> a working path. The recovery<br />

(alternative) path is not used. The connection is interrupted when any <strong>of</strong> its<br />

elements fails. Thus, its availability equals simply an availability <strong>of</strong> a path on<br />

which this connection is established. As a path we consider a set <strong>of</strong> #L links <strong>and</strong> #N<br />

nodes through which a signal from an input node (source) to an output node (sink)<br />

is transmitted. From the reliability st<strong>and</strong>point, this is a serial structure, i.e., the<br />

structure remaining in the operation state only if all <strong>of</strong> its elements are in the up<br />

state. Therefore, the availability <strong>of</strong> a path ( A ) is determined as [4]:<br />

A<br />

=<br />

# L<br />

∏<br />

ALi N j<br />

A<br />

P<br />

×<br />

# N<br />

∏<br />

A<br />

P Li<br />

N j<br />

i=<br />

1 j=<br />

1<br />

where is the availability <strong>of</strong> link i; A is the availability <strong>of</strong> node j.<br />

3. DEDICATED PROTECTION<br />

The simplest protection scheme is called dedicated protection or 1+1<br />

protection. From the reliability point <strong>of</strong> view it is a parallel-serial structure with hot<br />

st<strong>and</strong>by, because traffic is carried both through the working <strong>and</strong> the protection<br />

paths all the time. The parallel structure means that only if all N paths fail, the<br />

connection is down. Availability for the parallel structure is calculated according to<br />

the following formula [5]:<br />

(1)


Availability Assessment <strong>of</strong> Resilient Networks<br />

N<br />

∏<br />

i= 1<br />

A(parallel item1, …, itemN) = 1 − P (itemi<br />

= unav) = 1 − ( 1 − A ) (2)<br />

where P(event) is the probability <strong>of</strong> an event; ‘unav’ means that an item is<br />

unavailable; Ai is the availability <strong>of</strong> item i.<br />

Therefore, the availability <strong>of</strong> the whole connection using the 1+1 protection<br />

( A1+<br />

1 ) will be calculated as follows (see Fig. 1):<br />

ASo ASp<br />

( ( AP<br />

)( AP<br />

) ASw<br />

A<br />

1 2 Si<br />

1 1 1<br />

A1+ 1 = − − −<br />

(3)<br />

where , A are the availabilities <strong>of</strong> the source <strong>and</strong> sink node, respectively;<br />

ASo Si<br />

APi P2<br />

is the availability <strong>of</strong> the first path, <strong>and</strong> <strong>of</strong> the second path; , A are the<br />

availabilities <strong>of</strong> a splitter <strong>and</strong> switch, respectively.<br />

N<br />

∏<br />

i=<br />

1<br />

A ASp Sw<br />

Using the example <strong>of</strong> a Unidirectional Path-Switched Ring (UPSR), a<br />

particular case <strong>of</strong> the 1+1 protection, we present how practically calculate the<br />

availability. The availability <strong>of</strong> connections established in some <strong>of</strong> ring topologies<br />

is relatively simple to calculate; some effort has already been made [6]. In UPSR<br />

each pair <strong>of</strong> nodes is connected by two separate paths. Traffic from the source node<br />

is transmitted simultaneously in both directions: clockwise on the working path <strong>and</strong><br />

counterclockwise on the protection path. The working path is located on a fiber<br />

which is different from the fiber carrying the protection path. In the case <strong>of</strong> any<br />

working path element failure, the sink node starts to receive the signal from the<br />

protection path [7].<br />

Figure 1. Dedicated protection: reliability block diagram<br />

If we assume that availabilities <strong>of</strong> all nodes are equal <strong>and</strong> each span is <strong>of</strong> an<br />

identical length (which means that also availabilities <strong>of</strong> all spans are equal) the<br />

availability <strong>of</strong> a connection between two nodes in UPSR ( A ) is given by the<br />

UPSR<br />

following formula (cf. the reliability block diagram shown in Fig. 2):<br />

A A 1−<br />

1−<br />

A 1−<br />

A A<br />

( ( )( ) ) =<br />

UPSR = N<br />

WP PP N<br />

k k −1<br />

n n−1<br />

⎛ ⎛<br />

⎞⎛<br />

⎞⎞<br />

= ⎜ − ⎜1−<br />

∏ A S × ∏ AN<br />

⎟⎜1−<br />

∏ A × ∏ ⎟<br />

⎜<br />

S ⎟<br />

⎟<br />

=<br />

⎝ ⎝ i=<br />

1 i=<br />

1 ⎠⎝<br />

i=<br />

k + 1 i=<br />

k + 1 ⎠⎠<br />

2 A N 1 AN<br />

2<br />

= A<br />

k k −1 1−<br />

1−<br />

A A<br />

n−k<br />

n−k<br />

−1<br />

1−<br />

A A<br />

n−k<br />

n−k<br />

+ 1 k k + 1<br />

= A A + A A − A<br />

N<br />

n n<br />

( ( )( ) A<br />

S<br />

N<br />

S<br />

N<br />

S<br />

N<br />

S<br />

N<br />

S<br />

i<br />

N<br />

(4)


Piotr Chołda, Andrzej Jajszczyk<br />

where A , , , A are the availabilities <strong>of</strong> a node, a span, the working <strong>and</strong><br />

N<br />

AS AWP PP<br />

the protection path, respectively; k is the number <strong>of</strong> spans between source <strong>and</strong> sink<br />

nodes; n is the total number <strong>of</strong> spans in the ring.<br />

The availability function achieves its minimum in the case <strong>of</strong> the connection<br />

between the most distant nodes (the nodes which are present at the opposite sides<br />

<strong>of</strong> the ring). Fig. 3 illustrates the way the availability is changing depending on the<br />

length <strong>of</strong> the protection path ( A A = 99.<br />

95%<br />

, n = 16 ). The graph is<br />

N<br />

= S<br />

symmetric: the availabilities calculated for connections which can be characterized<br />

by span length k equal values calculated for connections with span length n − k .<br />

It also means that the connection from X to Y has the same availability as the<br />

connection from Y to X.<br />

Figure 2. Reliability block diagram for a connection established in UPSR<br />

Figure 3. A comparison <strong>of</strong> connection availabilities in UPSR<br />

4. SHARED PROTECTION<br />

Except for the 1:1 protection case, that from the reliability st<strong>and</strong>point, has the<br />

same structure as the 1+1 protection, schemes describing shared protection form


Availability Assessment <strong>of</strong> Resilient Networks<br />

non-series-parallel structures (an example: Fig. 4a). For exact calculations,<br />

methods based on enumeration <strong>of</strong> probabilistic events should be used. We employ<br />

a simplified model which reduces all switches to one (see Fig. 4b). Therefore, it<br />

will be possible to apply calculations for the r-out-<strong>of</strong>-s reliability structure (i.e., the<br />

whole system is up if r out <strong>of</strong> s items are in the operating state). We will calculate<br />

the availability <strong>of</strong> the 1:N protection scheme ( ), i.e., for N-out-<strong>of</strong>-(N+1)<br />

A 1:<br />

N<br />

reliability structure. In such a method one protection path is shared by N working<br />

paths. For clarity, we disregard unnecessary details such as the source/sink nodes<br />

availability (as is commonly practiced [8]). The following formula considers the<br />

only availability <strong>of</strong> the “parallel” part:<br />

N N<br />

N + 1<br />

N + 1<br />

⎛ ⎞<br />

A ∏ ∑⎜( ) ∏ ⎟<br />

1:<br />

N = Ai<br />

+<br />

⎜<br />

1 − Ai<br />

Aj<br />

⎟<br />

= ∑∏ Aj<br />

− N∏<br />

Ai<br />

(5)<br />

i = 1 i= 1 ⎝<br />

j≠i<br />

⎠ i= 1 j ≠i<br />

i = 1<br />

where is the availability <strong>of</strong> path i ( A for the protection path).<br />

Ai N + 1<br />

While calculating Eq. (5) we take into account N+1 probabilistic events: one in<br />

which all <strong>of</strong> N working paths are in the up state, <strong>and</strong> N similar events in which one<br />

(in each event different) <strong>of</strong> the working paths fails, but all remaining paths<br />

(including the protection one) are up. The availability calculated according to the<br />

presented method concerns the probability that the whole protected traffic will be<br />

carried properly. We have to remember, however, that even if some working paths<br />

<strong>and</strong> the protection path fail, the rest <strong>of</strong> the traffic (i.e., that on working paths in the<br />

up state) will be transmitted. Therefore, we consider the best case.<br />

Figure 4. 1:N shared protection: a) exact view <strong>of</strong> the structure, b) simplified structure


Piotr Chołda, Andrzej Jajszczyk<br />

As a numerical example, we compare three different protection topologies:<br />

three working connections with the same availabilities are to be protected. The<br />

following schemes are compared: the 1:3 protection; the topology in which each<br />

working path has its own protection path, it is in fact the 1+1 protection applied<br />

three times (six paths in total); <strong>and</strong> the 3:3 protection (three protection paths shared<br />

by three working paths). The analytical formula for the last case is rather complex,<br />

so we do not present it here. It is based on the 3-out-<strong>of</strong>-6 structure. The relevant<br />

formulas <strong>and</strong> comprehensive information how to calculate them can be found in<br />

[9]. The results are presented in Fig 5. We can see that the 1:3 protection results in<br />

a large increase <strong>of</strong> the availability in relation to unprotected connections. The gain<br />

is especially visible in the case when availabilities <strong>of</strong> the working paths are small<br />

(even if availabilities <strong>of</strong> protection paths are relatively low). It is the least reliable<br />

but the most cost efficient scheme <strong>of</strong> the three compared. Moreover, from the<br />

reliability point <strong>of</strong> view, the most favorable is the 3:3 protection, even better than<br />

applying the 1+1 protection for each working path separately. The reason is that in<br />

the case <strong>of</strong> 3:3 protection the scheme is more flexible. When one <strong>of</strong> the working<br />

paths fails, the traffic is rerouted on the protection path. And again, when this path<br />

fails, the traffic can be rerouted again. It is impossible in the case <strong>of</strong> the 1+1<br />

protection. However, we have to remember that the gain is minimal <strong>and</strong> practically<br />

such a solution could not be advantageous: it is, after all, necessary to use<br />

additional switching elements. Some analyses [10] indicate that the influence <strong>of</strong><br />

switching nodes cost on the total expenditure can be quite large. It seems that this<br />

fact is <strong>of</strong>ten neglected <strong>and</strong> optimization based only on the link cost could be<br />

insufficient.<br />

5. RESTORATION<br />

Unlike in the case <strong>of</strong> a protection, where an alternative path is established<br />

before a failure, in the case <strong>of</strong> the restoration such a path is sought after the fault<br />

notification. It makes the calculation <strong>of</strong> the availability more difficult. For example,<br />

the availability is calculated not for connections, but for carried traffic. Therefore,<br />

it is called the availability <strong>of</strong> a load ( A ). The authors <strong>of</strong> [4] propose the<br />

following formula for calculating it in a restoration scheme:<br />

M ⎛ CRec<br />

⎞ 1<br />

A = − ∑ ( ) × ⎜ ⎟<br />

Load 1 P Sceni<br />

⎜<br />

1−<br />

⎟<br />

(6)<br />

i= 1 ⎝ CT<br />

⎠<br />

where P ( Sceni<br />

) is the probability <strong>of</strong> failure scenario i (out <strong>of</strong> M scenarios); C T is<br />

the total capacity <strong>of</strong> traffic (sum <strong>of</strong> all working capacities); is the total<br />

capacity <strong>of</strong> the connections recovered upon failure scenario number i (this capacity<br />

includes also capacities <strong>of</strong> connections unaffected by a failure).<br />

Load<br />

C Rec<br />

i


Availability Assessment <strong>of</strong> Resilient Networks<br />

Eq. (6) is very useful. Firstly, it enables us to calculate the availability <strong>of</strong> a<br />

load when no restoration procedures are applied. Consider the following example:<br />

we have three physical links: how to allocate three connections on them to reach<br />

the highest availability? There are no protection schemes, so in the case when link i<br />

(this link has capacity C ) fails, the recovery capacity which can be used by<br />

traffic carried by link i is:<br />

Wj<br />

= ∑<br />

j≠<br />

i<br />

C C . P ( Scen ) is the probability that a link<br />

Reci<br />

will fail, so it is equal to the unavailability <strong>of</strong> link i: ( i ) i<br />

Wj<br />

i<br />

P Scen = 1−<br />

A . The<br />

availability <strong>of</strong> a load is an average <strong>of</strong> path availabilities weighted by capacities.<br />

The calculation results are presented in Table 1. The connection which carries the<br />

greatest load brings the largest contribution. If one pays attention to rows printed in<br />

bold, will be able to notice that it is better to increase the availability <strong>of</strong> such a<br />

connection (with capacity <strong>of</strong> 25 units) than to even more significantly increase the<br />

availability <strong>of</strong> a connection carrying much less capacity (10 units).<br />

Figure 5. A comparison <strong>of</strong> three protection schemes<br />

The restoration will be considered only for a very simple case: we will model<br />

some procedures related to MultiProtocol Label Switching. Each Label Switched<br />

Path (LSP) i between two Label Switching Router (LSR) nodes is characterized by<br />

the three parameters: its availability (Ai), capacity <strong>of</strong> LSP in the up state (CWi) <strong>and</strong><br />

capacity reserved for restoration on the link traversed by the path (CPi). In the case


Piotr Chołda, Andrzej Jajszczyk<br />

<strong>of</strong> restoration the data from an affected LSP should be divided into some recovery<br />

sub-paths, i.e., we deal with the bifurcated recovery paths. When a path fails,<br />

restoration is based on the allocation <strong>of</strong> all available capacity reserved on all other<br />

paths. In some cases less capacity than that used by working LSP before failure<br />

will be allocated to recovery sub-LSPs. We take into account not only single<br />

failures but also double failures. In the latter case (when physical paths i <strong>and</strong> j fail)<br />

P Scen = 1−<br />

A 1−<br />

A .<br />

the probability <strong>of</strong> such a scenario equals ( ) ( )( )<br />

Table 1. Comparison <strong>of</strong> availabilities <strong>of</strong> a load<br />

Capacity 10 15 25 Availability <strong>of</strong> a load<br />

99,95% 99,90% 99,85% 99,89%<br />

99,90% 99,95% 99,85% 99,89%<br />

99,95% 99,85% 99,90% 99,90%<br />

Availability<br />

<strong>of</strong> a path<br />

99,85% 99,95% 99,90% 99,91%<br />

99,99% 99,85% 99,91% 99,91%<br />

99,90% 99,85% 99,95% 99,91%<br />

99,85% 99,90% 99,95% 99,92%<br />

We considered the problem how to allocate three different LSPs on three<br />

physical links (cf. Table 2 for data concerning the example; total capacity reserved<br />

for restoration equals total working capacity, i.e., 6 units). We checked all possible<br />

situations <strong>of</strong> allocating LSPs <strong>and</strong> reserved capacity on different physical links (Eq.<br />

(6) was applied again). Some typical examples are presented in Table 3. We can<br />

see that the most favorable solution is to allocate paths in such a way that the path<br />

carrying the largest working capacity is allocated on the physical link with the<br />

largest availability, <strong>and</strong> simultaneously to allocate the reserved capacity on the link<br />

which has the second-largest availability (cf. the last row <strong>of</strong> Table 3). On the other<br />

h<strong>and</strong>, the use <strong>of</strong> the worst link (from the reliability st<strong>and</strong>point) is not<br />

recommended, unless one is forced to do it (e.g., all capacity on “better” links is<br />

already used). Moreover, it would be reasonable to allocate the reserved capacity<br />

(instead <strong>of</strong> the working capacity) on the worst link (characterized by the lowest<br />

value <strong>of</strong> availability). We can also note that if it is not necessary it would be<br />

favorable not to balance utilization <strong>of</strong> paths (e.g., not to allocate paths in the<br />

following way: LSP 1 — link 1, LSP 2 — link 2, etc.), but allocate it in an “unjust”<br />

way (e.g., no capacity on link1; LSP 2 <strong>and</strong> LSP 3 on link 3, <strong>and</strong> the rest <strong>of</strong> capacity<br />

on link 2). Finally, we notice that the capacity reserved for restoration should be<br />

allocated on links different from those carrying the working capacity. Generally,<br />

the larger amount <strong>of</strong> working capacity is carried by a link, the smaller reserved<br />

capacity should be carried on it.<br />

ij<br />

i<br />

j


Availability Assessment <strong>of</strong> Resilient Networks<br />

Table 2. Restoration example: used values<br />

LSP 1 LSP 2 LSP 3<br />

Capacity 1 unit 2 units 3 units<br />

Physical link 1 Physical link 2 Physical link 3<br />

Availability 99,85% 99,90% 99,95%<br />

Table 3. Example allocations <strong>of</strong> capacities <strong>and</strong> calculated availabilities<br />

A1 Cw1 Cp1 A2 Cw2 Cp2 A3 Cw3 Cp3<br />

Availability <strong>of</strong> a load<br />

without restoration<br />

Availability <strong>of</strong> a load<br />

with restoration<br />

99,85% 5 6 99,90% 1 0 99,95% 0 0 99,86% 99,87473%<br />

99,85% 5 6 99,90% 0 0 99,95% 1 0 99,87% 99,87473%<br />

99,85% 5 5 99,90% 1 0 99,95% 0 1 99,86% 99,89975%<br />

99,85% 0 0 99,90% 2 6 99,95% 4 0 99,93% 99,96657%<br />

99,85% 5 2 99,90% 1 0 99,95% 0 4 99,86% 99,97483%<br />

99,85% 3 0 99,90% 2 2 99,95% 1 4 99,88% 99,99990%<br />

99,85% 0 2 99,90% 3 2 99,95% 3 2 99,93% 99,99992%<br />

99,85% 0 2 99,90% 3 3 99,95% 3 1 99,93% 99,99992%<br />

99,85% 0 0 99,90% 3 3 99,95% 3 3 99,93% 99,99995%<br />

99,85% 0 0 99,90% 2 4 99,95% 4 2 99,93% 99,99995%<br />

99,85% 0 0 99,90% 1 5 99,95% 5 1 99,94% 99,99995%<br />

6. CONCLUSION<br />

Our intention was to show how simple models <strong>of</strong> protection schemes enable<br />

fairly useful evaluation <strong>of</strong> the basic reliability measure, i.e., the availability.<br />

Although the models are approximate, they can show us some necessary hints<br />

concerning network design. It is <strong>of</strong> a great importance to note that we focus only on<br />

availability problems <strong>and</strong> neglect other issues, for example cost-effectiveness. We<br />

can note that some solutions are characterized by a high availability, but they are<br />

not economical. The problem <strong>of</strong> balancing reliability parameters <strong>and</strong> cost requires<br />

introduction <strong>of</strong> some more complex mathematical models.<br />

BIBLIOGRAPHY<br />

[1] Clouqueur M., Grover W. D.: Computational <strong>and</strong> Design Studies on the Unavailability<br />

<strong>of</strong> Mesh-restorable Networks, Proceedings <strong>of</strong> IEEE/VDE Design <strong>of</strong> Reliable<br />

Communication Networks (DRCN 2000), Munich, April 2000, pp. 181-186<br />

[2] Grover W.: Mesh-Based Survivable Networks. Options <strong>and</strong> Strategies for Optical,<br />

MPLS, SONET, <strong>and</strong> ATM Networking, Prentice Hall PTR, Upper Saddle River, New<br />

Jersey 2004<br />

[3] ITU-T Recommendation E.800: Terms <strong>and</strong> definitions related to quality <strong>of</strong> service <strong>and</strong><br />

network performance including dependability, August 1994


Piotr Chołda, Andrzej Jajszczyk<br />

[4] Willems G., Arijs P., Van Parys W., Demeester P.: Capacity vs. Availability Trade-<strong>of</strong>fs<br />

in Mesh-Restorable WDM Networks, Proceedings <strong>of</strong> 3 rd Int. Workshop on the Design <strong>of</strong><br />

Reliable Communication Networks (DRCN 2001), Budapest, Hungary, October 2001<br />

[5] De Maesschalck S.et al.: Pan-European Optical Transport Networks: an Availabilitybased<br />

Comparison, Photonic Network Communications, Vol. 5, No. 3, May 2003, pp.<br />

203-225<br />

[6] Liew S. C., Lu K. W.: A Framework for Characterizing Disaster-Based Network<br />

Survivability, IEEE Journal on Selected Areas in Communications, Vol. 12, No. 1,<br />

January 1994, pp. 52-58<br />

[7] Ramasvami R., Sivarajan K. N.: “Chapter 10: Network Survivability”, in: Optical<br />

Networks: A Practical Perspective, Morgan Kaufmann, San Francisco 2002, California,<br />

USA, pp. 537-589<br />

[8] Held M., Wosinska L., Nellen P. M., Mauz Ch.: Consideration <strong>of</strong> Connection<br />

Availability Optimization in Optical Networks, Proceedings <strong>of</strong> 4 th Int. Workshop on the<br />

Design <strong>of</strong> Reliable Communication Networks (DRCN 2003), pp. 173-180 , Banff,<br />

Alberta, Canada, October 2003<br />

[9] Grosh D. L.: A Primer <strong>of</strong> Reliability Theory, John Wiley & Sons, New York 1989<br />

[10] Lakatos Zs., Jakab T.: Protection or restoration: a generic study on the impact <strong>of</strong> line<br />

costs <strong>and</strong> switching, Proceedings <strong>of</strong> 3 rd Int. Workshop on the Design <strong>of</strong> Reliable<br />

Communication Networks (DRCN 2001), Budapest, Hungary, October 2001

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!