Abstract—Passive Optical Networks (PONs) are one of the preferred technologies to deploy broadband access networks. As time passes, end users presuppose network connectivity to be always available, and expect PONs to be highly dependable. Yet operators, from an economic view, are interested in the costs related to failures. Thus, PONs dependability and associated costs have been extensively studied, but only focusing on hardware failures. Contrarily, this paper performs a thorough analysis of the impact of software failures in failure-related costs.
Based on real empirical data, software failures are thoroughly characterized and classified in four different categories according to their severity. Also, the effect of software failures on the behavior of PON’s fiber protection and recovery mechanisms is detailed. Software failures are included into a Markov cost model, implementing a comprehensive cost framework. This way, the dependability-related costs of PONs are analyzed, accounting for hardware and software failures, as well as for the consequences of software failures on well-known PON protection mechanisms. Moreover, how the testing phase duration and user profile (residential or business) impact these costs is pinpointed.
Index Terms—Failure coverage; Operational Expenditures;
Passive Optical Networks; protection; software failures
I. INTRODUCTION
Due to the rapidly increasing bandwidth demands of new services, network operators are being pushed towards the deployment of broadband access networks. Certainly, Passive Optical Networks (PONs) are widely recognized as the best suited solution for supporting such demands [1]. Amid other features, PONs offer high bandwidth on a per-user basis, as well as scalability and flexibility. PONs also present low energy consumption and are cost-effective as costs are shared among several customers. Hence, PONs and Next-Generation PONs (NG-PONs) are regarded as the most promising solution for future fiber-based access networks [2].
Yet, as end users are starting to take network connectivity for granted, dependable service delivery is also expected from PONs. Consequently, to satisfy the need of reliable access, dependability of access networks has become an important case of interest nowadays. In fact, several protection schemes and dependability analyses for different PONs and NG-PONs flavors can be found in current literature [3], [4], [5].
Commonly, a system’s dependability is assessed by its
A. Fernández and N. Stol are with the Department of Telematics, Norwegian University of Science and Technology, Trondheim, Norway (e- mail: alvarof@item.ntnu.no).
availability. Still, from a financial point of view, an operator is typically more interested in the failure-related costs, known to be part of the operational expenditures (OPEX). Notably, this interest arises as a proper understanding of failure-related OPEX can be used in cost optimization or risk management analyses. Usually, failure-related OPEX cover the cost of repair and extra equipment, payment of penalties and loss of reputation if a large number of users are affected by failures.
However, most of the published PON dependability studies are focused only on hardware, physical faults. Few papers address software dependability or its consequences with respect to OPEX, even though software faults account for an important part of service failures in many systems [6].
Furthermore, software failures also represent impairments to the correct behavior of protection schemes. This is more important as PONs/NG-PONs evolve in complexity, serve more users or are used in e.g. data centers.
Chiefly, this paper provides a comprehensive analysis of the effects of software failures in Time Division Multiplexed (TDM) PONs’ failure-related OPEX. Extending the work in [7], a thorough characterization of Gigabit-capable TDM PONs (GPONs) software failures is performed, based on empirical data [8], [9]. How software failures hinder the performance of fiber protection schemes (i.e. fault coverage) in TDM PONs is also deeply detailed, based on real data [10].
Applying Duane model for software reliability growth [11], the software failure intensity is estimated as a function of the testing time and included in a Markov cost model. Hence, the impact of hardware and software failures, as well as of imperfect fiber protection recovery (due to software) in PON’s failure-related OPEX is analyzed, accounting for the length of the testing phase and the user profile (residential or business).
This paper is organized as follows. First, Sect. II presents the PON architecture and fiber protection scheme. Section III describes the software dependability and failure coverage modelling. Section IV details the Markov cost model used to assess the failure-related OPEX, while results are presented in Sect. V. Finally, Sect. VI gives the conclusions of this work.
II. TDMPONARCHITECTURE AND PROTECTION
Plainly, the basic TDM PON architecture is shown in Fig. 1 (a). At the operator’s Central Office (CO), the Optical Line Terminal (OLT) is located, consisting of two elements: OLT ports where fibers are connected and the OLT chassis housing them. Besides, the OLT chassis also hosts the OLT software in charge of the PON correct behavior. At the user’s side, an
Influence of Software and Hardware Failures with Imperfect Fault Coverage on PONs OPEX
Álvaro Fernández, Norvald Stol
Op an pl ho Fo sp tw km an Fe Co be m m fe fe Su pr op wo pa (d op ke Ne th
Fig
ev re pr tes fa wh [1 on sh (ta
ptical Networ nd the ONUs, t aced in a stre ousing a set ollowing the plit ratio of 1:
wo sections of m [12]): Feed nd Distribution
eeder fibers sp ontrarily, the l
Concerning p een shown to mechanisms by mainly due to th
eder fiber fa eder fiber cu uccinctly, this rotection feede ptical switch a orking and pr aths/trenches t digging), loss ptical switch w eep connectiv ecessarily, the e switching an
g. 1. Basic unprot
III. OLT Essentially, valuated by
liability grow redicting the so st and debugg ilures occur ac hose failure in
1] for a full de Notably, th n the analysis hown to fit e aken from [8
rk Unit (ONU the Remote N eet cabinet. T of the passiv GPON ITU-T :32 are assum f fiber can be
er Fibers (FF) n Fibers (DF) pan several ki length of distr protection in P
be one of th y several autho
he large numb ailure and the ut (as feeder f mechanism i er fiber betwe at the CO, as rotection feed to avoid comm
of signal will will flip to the vity between e OLT softwa nd preserve se
tected (a) and feed
TSOFTWARE D the dependab means of th wth. Reliabil
oftware failure ging processe
ccording to an ntensity decrea escription of t he OLT softw presented in empirical data ]). Thus, the
U) is deployed Node (RN) acts The RN consis ve elements f T standard [1 med as passive
identified (ma ), between the , between the ilometers and ribution fibers PONs, feeder he most cost- ors [3], [4], [5 ber of end user e relatively la
fibers cover s implies the de een the OLT a s shown in Fi
er fibers mus mon failures.
l be detected protection fee n the OLT are should be p ervice provisio
der fiber protecte
DEPENDABILIT
bility of the he Duane m lity growth e intensity tak s. In the Dua n inhomogene ases with the t this model).
ware modelling [7], where the a from a GP
e software fa
d. Between th s as a splitting sts of a RN c for signal spl 12], splitters w
e elements. F aximum reach e OLT and th RN and the O serve all end is usually sma fiber protectio -efficient prot 5]. Decidedly, rs affected in c arge probabil several kilom eployment of and the RN, a ig. 1 (b). Cer
t span over d In case of fib at the CO, an eder fiber in or
and the O prepared to pe on after it.
d (b) PON archite
Y MODELLING
OLT softw model for so models allow king into accou ane model, so
ous Poisson p testing time (s g in this work e Duane mode PON OLT so ailure intensity
he CO g point,
chassis litting.
with a Finally, h of 20 he RN;
ONUs.
users.
aller.
on has tection this is case of lity of meters).
both a and an rtainly, disjoint ber cut nd the rder to ONUs.
erform
ecture.
G
are is ftware w for unt the
ftware process ee e.g.
builds el was ftware y as a
func
t zY clas perf clea In t clas
“ave high prec (reg repo (bas S (10%
caus trivi S [8]
Wh Con serv Res S (31%
the Still (bus S (2%
defe serv A Mar A.
F cov defi succ mec whi In prot Chi GPO swit soft B fibe Surp corr prot the busi
ction of the tes
0.237133 Yet, the workssification and forming a mo arer indication the empirical ssified into fo erage critical hly critical bu cise, it marke garding severi orted in [8]
sed on real pro Severity 4 (S4
% of total b sing service d ial errors not l Severity 3 (S3 (57% of tota ereas services nsequently, S3 ved by the OL sidential users Severity 2 (S2)
% of total bug services are st l, no total sto siness and resi Severity 1 (S1
% of total bugs ects cause tota ved by the OL Additionally, th
rkov model is Feeder Fiber Further extend erage is now ined as the pr ceeding upon chanism does ich requires ad n order to estim tection, the r
efly, [10] des ON deployme tching were em tware failures Based on the er protection
prisingly, onl rect service r tection fault c cases medium iness users a
sting time (t) w .
*t0.238842 k in [7] adopt d description. C
ore thorough n of failure con
data reported our categories l bugs”, “hig ugs”. Althoug edly matches
ity) from [9].
will be mode ojects), which 4) defects refe bugs – 0.1 p disruption. Fo leading to failu ) defects map al bugs – 0.
s are not interr 3 defects affe LT (due to b
(without strin ) defects relate gs – 0.31 pro topped and pe oppage occurs idential) are re ) defects refe s – 0.02 proba al stoppage, al T cease being he exact mapp
detailed in Se Protection Fa ding the work w introduced.
robability of t n failure occu s not succeed dditional recov mate the fault results reporte cribes a fault ent, where fe mulated. Afte caused service se results, th
mechanism ly 20% of th restoration. C coverage is fix m criticality fa
are not corre
was found in [
ted a minima Consequently, failure charac nsequences on d in [8], softw s; namely “lo ghly critical b
gh this bug d the software . Purposely, t elled followin
assumes four er to “low pri probability of or example, c ure fall in this p into “averag 57 probability rupted, perform fect a subset bandwidth and ngent demands e to “highly cr obability of oc erformance is h
s, thus only a egarded as dow er to “very hig
ability of occu ll business an g served.
ping of softwa ect. IV.
ault Coverage k in [7], the
In dependab the failure rec
urrence [13].
d, an uncove very actions.
t coverage rega ed in [10] w injection cam eeder fiber cu er restoring co
es not to be co e fault cover can be estim he emulated Consequently, xed as 0.2. B ailures (mappe ectly restored
[7] to follow
al software fa , this paper aim cterization, wi n service deliv ware failures w ow priority bu
bugs”, and “ description is defects taxon the software ng this taxon severity level iority bugs” in f occurrence),
cosmetic fault s category.
ge critical bug y of occurren mance is hinde of business u d delay deman s) are not affec ritical bugs” in ccurrence). Pa
heavily hampe a number of u
wn.
ghly critical b urrence). As t nd residential u are failures into e
concept of ility, this term covery mechan
. If the reco ered fault oc
arding feeder will be emplo mpaign carried ut and protec nnectivity, sev orrectly recove age of the fe mated as foll cases resulte the feeder Besides, in 71%
ed as S3 defec d) were repo
(1) ailure ms at ith a very.
were ugs”,
“very s not nomy
bugs nomy
ls.
n [8]
, not ts or s” in nce).
ered.
users nds).
cted.
n [8]
art of ered.
users bugs”
these users o the
fault m is nism overy curs, fiber oyed.
d in a ction veral ered.
eeder lows.
ed in fiber
% of cts – orted.
He fib cr th pr Fi fe sa co re co ar
I m so no m th (fa m th ar tra Cij
eq Co as pa an PC (S th of of sim an bu an
$/
ou is P
fa wi sp ar de Be
ence, the prob ber failure is riticality failur
e affected us robability of inally, these p
eder fiber fai ame proportion We would li overage of 0.2 sults, it was n overage should re better seen a IV. MARKOV
In order to an model has been
oftware failure o software fail model is especi
e software-ha ault coverage models (reliabil Briefly, two is study, name re associated ansition from
ij (in $). Mo quipment that
onsequently, i Additionally, ssociated to e ayment of the nd payment o CR). Simply, t SH and SS for h
e number of re As for the PC f reputation du f revenue ins
mplicity. Hen nd PRB in $/h usiness users) nd a reputation /h). To accoun utages occur (e introduced (χR
(
,B R K
i F
PCR
Let us consid ilures. State d ith the PON e plitting ratio, 3 re taken from epends on the esides, it is a
bability of an u 0.71. Finally res (correspon sers are not r a S2 uncover probabilities ( ilures) will va n as the softwa ike to remark 2 is unexpecte not possible to
d be taken wit as a worst case
V COST MODEL
nalyze the TD n employed [1 es have been i lures is used ially well suit ardware inter ) in Sect. III lity blocks) du
types of failu ely cost impul
to transition state i to state ore precisely must be boug impulse costs o , cost rates (c each state i. C
repairmen (d of penalties (r
the RCR in a s hardware and epairmen in st CR, it includes ue to failures.
tead of cost, ce, the PCR d h, subscripts , the failed cl n rate gauging nt for an incr e.g. negative p χR and χB). Thu
,i * K
K RR
FCK
der the Marko definition depe elements descr 32 ONUs are m [3] and [15]
fiber length, assumed that t
uncovered S3 y, 9% of the nding to S2 de
recovered). C red feeder fib (which apply ary with the are failure inte that to our a dly low. Due verify it. Sub th care. Thus, e scenario.
L FOR FAILURE
DM PON’s OP 14]. As in [7],
included. Thu as baseline. A ed for this wo raction and
cannot be m ue to independ ure-related cos lses and cost r s in the Ma e j has an asso y, these cost ght to replace only apply to
i – cost per u Cost rates con denoted Repair
referred as Pe state i is propo d software rep
tate i (OCi).
s the cost of pe Yet the latter
it is include depends on the R and B de lients in state g reputation co reased cost of press releases) us, the PCR in
,i* K
K PR
FC
ov cost model ends on the typ
ribed in Sect.
assumed. Fail ]. Notably, th allowing for there is only
failure upon cases caused efects – a num orrespondingl ber failure is only to unco testing time ensity (1).
appreciation, a to the lack of bsequently, thi
results in this
E-RELATED OP PEX, a Marko , both hardwa us, the scenario Also, a Marko
ork. This is be imperfect rec modelled with dence assumpti sts are conside rates. Cost im arkov model,
ociated impuls ts cover the
a faulty comp hardware failu unit time in $/
nsist of two r Cost Rate – enalty Cost R ortional to the pairmen, in $/h
enalties and th r can be seen a ed in the PC e penalty rates enote resident e i (FCR,i and ost (RRR and R
f reputation if ), an impact fa state i follows
K).
with only har pe of failed ele II. Due to th lure and repai he fiber failur different scen one repair cr
feeder d high mber of
ly, the s 0.09.
overed in the a fault f other is fault s work
PEX ov cost are and o with ov cost ecause covery h static
ions.
ered in mpulses i.e. a se cost
extra ponent.
ures.
/h) are terms:
RCR) Rate –
salary h) and he cost as loss CR for
s (PRR
tial or FCB,i) RRB in f large actor χ
s:
(2) rdware ement, he 1:32 ir rates re rate narios.
rew. If ther the C also Acc chas clien affe faile and digg disc occu Like clien are A.
C exte com do n OLT occu seve soft failu repa In testi den and
A S1 As m user vari for F resp serv
Fig.
re are two or highest reduct Concerning the o depends on cording to pre ssis affect 16 nts. OLT por ect only 1 clien
ed clients is m d 5000 clients
ging close t connecting a urs far from th ewise, the uni nts for a distr uniformly dist Software Fail Concisely, soft ending the mo mbinations are
not lead to fa T software is ur. Yet, low erity software tware runs on ures if the O air of the OLT n Fig. 2, λsoft
ing time t. Fa oted λOLT_C an d S3 failures pr As for the num failures cause mentioned in rs (residentia iable from 1 a
S3 failures, bu Finally, softwa
pect to S1, S2 vice stoppage,
2. Markov model
more failed e tion in cost in e number of f n the type an evious studies 600 clients, rts and splitte nt. As for feed modelled as a
s. Intentionall to the CO w
large number he CO, the num
iform variable ibution fiber c tributed amon lures Modellin ftware failures odel as shown
not depicted ailed clients, working prop priority failu e failure has n the OLT ch OLT chassis h T chassis assum
ft denotes z(t) ailure and repa nd μOLT_C. Fina robability (0.0 mber of failed e total stoppag
Sect. III, S2 f al and busine
and 400. The ut only affectin are repair rates
and S3 failure a quick restar
l for OLT softwar
lements, the e a shorter time failed element nd number of [5], it is fixed while RN ch ers affect 32 c der fiber failur uniform varia ly, this mode
will cut sev r of clients.
mber of failed e is defined b cut. Business ng the total ser
ng
are included in Fig. 2 (all h for clarity). B they are not perly, any sev ures cannot a already occur assis, there ca has failed. Mo mes to fix soft
in (1), whic air rates of the ally p1, p2 and 02, 0.31 and 0.
clients due to ge, thus affec failures affect ess), modelle same uniform ng business us s are denoted es. As S3 failu rt (5 min., γ3 =
re failures.
element leadin e is repaired fi
ts in each sta f failed elem d as follows.
hassis affect clients and O
res, the numb able between els the fact th veral fibers, Yet if a dig d clients is sma between 1 and
clients (if pre rved clients.
in the analysi hardware-softw Because S4 fail
considered. If verity failures appear if a hi rred. As the annot be softw oreover, hardw tware failures.
h depends on e OLT chassi
p3 relate to S1 .57 respectivel o software failu
cting 1600 cli a subset of se ed as a uni m variable ap sers in this ran
γ1, γ2, and γ3
ures do not lea
= 12 h-1) where ng to irst.
ate, it ments.
OLT 100 ONUs er of 1000 hat a
thus gging aller.
d 100 sent)
is by ware lures f the
may igher OLT ware ware n the
s are 1, S2 ly).
ures, ients.
erved form plies nge.
with ad to e the
fa fa pr re B.
de pr sw fib ap fa ra cS
fa th rep
= un is If fa pr th af wo to
Fig
sim of th mu Th pr tra
E
iled process is ilures require rocessor is res load is assume Feeder Fibe To model the epicted in Fig revious states witch, the prot
ber failures ppropriate prob
il, the system Briefly, λFF ( ates of the feed
3 and cS2 are ilures, (0.71 a e fault cove pair rates are γ3 = 12 h-1 a niformly distri drawn. If both
the failure ilure, a secon revious result,
ese are the f ffect only busi orking fiber fa failed users. F
g. 3. Markov mod
V. DEPEN
In this sectio mulation, the f 1 year are pre
e expected c multiplying the he ECR is ass robabilities (p ansition from
(
i
ci
ECR
s reset is assu e a more co set and the s ed to take 30 m er Protection a e fiber protect g. 3 are added are not sho tection or the may be co babilities. If b is regarded as (μFF) and λSw
der fiber and e the probabi and 0.09 respe erage (correct the same as in and μUnc_S2 = ibuted random h fibers fail, th corresponds nd uniform v , is drawn. In failed clients;
iness clients i failures and pr Finally, switch
del for feeder fibe
NDABILITY-RE
on, after solv failure-related esented as exp ost over a gi Expected Co essed from th pi), and the im
state i to state )
*
j ij
Cij
umed to fix the omplex repair software and min. on averag and Fault Cov tion and fault d. For clarity, own. From a
working fiber overed or un both working a
s down.
(μSw) denote optical switch ilities of S3 ectively). Also t recovery).
n the previous γ2 = 2 h-1. U m variable betw
his is the numb to an uncove variable, betw n case of S2
whereas S3 in this range.
otection fiber h failures caus
er protection and f
ELATED OPEX ving the Mark d OPEX resul pected costs (in
iven time spa ost Rate (ECR he cost rate of mpulse costs
j (Cij and λij re .
*pi
e failure. S1 a r action wher data reloaded ge (γ1 = γ2 = 2 verage Modell coverage, the , combination fail-free stat r may fail. W ncovered wit and protection the failure (r h respectively.
and S2 unco o, 1- cS3- cS2 =
Uncovered fa s section, i.e.
Upon fiber fail ween 1000 and mber of failed c ered working ween 1000 an
uncovered f uncovered fa Recall that co failures do no se 32 failed cli
fault coverage.
XEVALUATION
kov cost mod lts over a time n $). In steady an is calculat R) by the time each state (ci) and rates of espectively) a
and S2 re the d. This 2 h-1).
ling e states
s with te, the orking th the n fibers
repair) . Also, overed
= 0.2 is failures μUnc_S3
lure, a d 5000 clients.
g fiber nd the failure, failures overed ot lead ients.
N
dels by e span y state, ted by e span.
), their f each as:
(3) In extr (RC to s Bes
A imp extr [3].
whe
$/h.
prot is a rate Yet whi are (χR) fibe scen feed resp leng R clien hard fibe inte with (h) (1).
sche I be s with perc OPE the failu sma (abo beg spot are redu imp
L the soft with fibe case the resp fact (due a 25
ntentionally, t ra equipment CR because of software failu sides, results a As for the para pulse costs, rel
ra equipment), The salary of ereas that of . Because resi tection, they a assigned to re e (discomfort t for business ile the reputat also different ) and 1.2 for b ers allow for narios model der and distr pectively. In s gths are fixed Results are pre
nts. As base dware failures er protection.
ended (making h software fai and the corres
The case la eme always w In Fig. 4, the seen that pen h no business centage of fai EX. Expected
software fail ures are prese all testing tim ove 40000 h inning of the tted and fixed present and a uction in OP portant the larg Let us focus n
OPEX. When tware failures h/without soft er failures (i.e es). Also, the
impact of sof pect to the im t that business e to bandwidth 5% of busines
the presented (regarding im f hardware fail
ures) and pen are presented w ameters, they lated only to h , the prices of f hardware rep software repa idential users are no subjecte sidential users with the oper users a penalt tion rate is 50 t for both typ business (χB).
modelling tw densely popul ribution fiber parse scenario to 18.2 and 1.
esented for di elines, the c s) are shown, w
If there is no g cS3 and cS2 i lures are pres sponding softw abeled as “FF works as intend results for de nalties accoun s clients. This
ilures affect s dly, increasing
lure rate) red ent. Although mes, is less im hours). Particu e testing phas d. After severa are more diffic EX due to i ger the percent
ow on the im n business cli in OPEX (i.e ftware) is big
e. comparing higher the per ftware failures mpact of feede s users are eas
h and delay de ss users, the i
d results are b mpulse costs) lures), softwar nalties (relati with 95% conf are defined a hardware failu f the compone pairmen (SH) i airmen (techni
are not willin ed to penalties s (PRR = 0), rator) is fixed ty of 100 $/h 0 $/h (RRB). T
pes, fixed to Finally, differ wo types of lated areas, w rs are 3.75 os (suburban o 8 Km. respect ifferent percen
ases with no with or withou o software, pr
in Fig. 3 equa ented for diffe ware failure in FProt.” assum ded (for referen
nse scenarios nt for most of
s is explained several client g the testing t duces the cos this reduction mportant for la ularly, this i se, software f al hours of te cult to identify
ncreased testi tage of busine mportance of so
ients are prese e. comparing
ger than the unprotected rcentage of bu s (the higher th er fiber failure ily affected by emands) justif impact of soft
broken down ), hardware re re repair (RCR ing to the P fidence interva as follows. Fo ures (purchasin ents are taken
is fixed to 190 icians – SS) i ng to pay extra s. Thus, no pen but the reputa d to 30 $/h (R is assumed (P The impact fac
1.1 for reside rent length fo scenarios. D with the length and 0.375 or rural areas) tively.
ntages of busi o software ( ut (“Unp.”) fe rotection work al to 0). The c ferent testing t
ntensity (h-1) mes the protec
nce purposes) are shown. It f the OPEX, d because a l ts, dominating time (i.e. redu sts when softw
n is noticeable arge testing t s because at failures are e sting, fewer f y and fix. Yet ing time is m ess users.
oftware failur ent, the impac
unprotected c impact of fe and FF prote usiness, the hi he difference es). Certainly y software fail fies this. Still, ftware failures
into epair R due CR).
als.
r the ng of from 0 $/h, is 80 a for nalty ation RRR).
PRB), ctors ential or the Dense hs for Km.
), the iness (only eeder ks as cases times from ction ).
t can even large g the ucing
ware e for times t the easily faults t, the more es in ct of cases eeder ected igher with y, the lures
with s and
fe ph fe so ca fe (e wi un im OP pr OP re tre do th su re fib as (fo les alw re
Fi
eder fiber fail hases. Also int eder fiber fa oftware failure ase, both effec Moreover, Fi eder fiber failu specially in b ith short fibe ncovered failu mportance of in
PEX analysis rotection with
PEX, even wi al analysis sho
Finally, Fig.
ends identified ominate the O e testing time ubstantially. E duction due to Necessarily, bers, making t ssociated to fe
or any percent ss relevant. A ways higher th levance as the
Decidedly, a
ig. 4. Expected co
lures becomes teresting is the ailures contri es, independen ts are similar f ig. 4 shows th ures do not co business-free ers, associate ures become ev
ncluding softw s. An idealiz hout software,
ith a high perc ows that this c
5 depicts the d before can b OPEX in spar
above 40000 Especially wit o increasing th the OPEX res this type of fa feeder fiber fa tage of busine Although the i han that of so e percentage o as feeder fibe
ost per client in de
comparable f e fact that, wit ibute more to ntly of the test
for short testin hat in dense sc ontribute exces areas). Main d failures ar ven rarer. Let ware failures i zed analysis, , may lead to centage of bu conclusion ma results in spar be also seen.
rse scenarios.
hours does no thout business he testing time
sults are high ailures more co
ailures now d ess), whereas s impact of fee oftware failure f business use ers are more
ense scenarios for
for very long t th no business o the OPEX ing time. Yet ng times.
cenarios, unco ssively to the O nly, this is be re uncommon
t us now rema in dependabili , e.g. feeder o almost neg usiness users. S ay be misleadin rse scenarios,
First, penaltie Second, incr ot reduce the O s users, the e is almost triv her due to the ommon. Thus dominate the O
software failur der fiber failu es, the latter ga ers increases.
prone to fa
r different percent
testing users, X than
in this overed OPEX ecause n, thus
ark the ity and fiber gligible Still, a ng.
where es also reasing OPEX OPEX vial.
larger s, costs OPEX res are ures is ains in
ailures,
unc not thes are Rea Km prof failu dep prot soft
A pres obje dire also sche for wor soft rese liter for cost app para reco
tages of business
overed failure very relevan se failures in t
included. This ach PONs (LG m.) have gaine
fitable evoluti ures will bec
endability an tection schem tware must be
A detailed failu sented in this ect of the stud ect impact of t o how they a
emes in PONs completeness rk include a d tware failures earch due to rature. Beside analyzing failu t models the
licable to an ameters, capt overy that cann
clients with varyi
es in dense sce t in a busine the OPEX is s result is espe G-PONs, exten d interest late ion of PONs. W come more l nd OPEX of mes. Thus, u taken into acc VI. CON
ure-related OP paper. Softwa dy, with a twof these failures i affect the pe s. Also, hardw . Based on rea detailed charac s in PONs, w the lack o es, a compreh ure-related OP eory has bee ny PON techn turing dynam
not be handled
ing testing time (s
enarios gain in ss-free scenar significant wh ecially import nding the fibe ely as a possi With this fiber likely to occ future PONs uncovered fai count and con
NCLUSIONS
PEX analysis are failures ha fold focus. Na in OPEX has erformance of ware failures h al data, the co cterization an which proves
f this inform hensive metho
PEX in PONs, en proposed.
nology by co mic interaction d by static mo
software failure in
n importance.
rio, the impac hen business u tant because L er reach up to ible economic r reach, uncov ur, hindering s even with ilures because
sidered.
of PONs has ave been the m amely, not only
been assessed f fiber protec have been inclu ontributions of nd classificatio useful in fu mation in cu od and framew
, based on Ma This metho orrect tailorin ns and impe odels.
ntensity).
Still ct of users Long- o 100 c and vered g the
fiber e of
been main y the d, but ction uded f this on of urther urrent work arkov od is ng of erfect
re bu fa pr of Re no sp it ot fa PO re lim PO th Si pr m in so ac
[1]
[2]
Fi
Decidedly, s lated OPEX o usiness users a ilures domina rotection, softw f the OPEX, in
egarding impe ot increase the parse deploym does contribu ther failures.
ilures should ONs. Also, i duces the eff mited for testin
Finally, this ON flavors an e imperfect re imilarly, dupli rotection. Yet may markedly h
teresting to as oftware failur ccompanied by
] F. Effenberge Commun. Mag ] G. Kramer, M optical access the IEEE, Vol ig. 5. Expected co
oftware failur of PONs, esp are present. Y ate the OPEX ware failures s ncreasing with erfect recovery e OPEX rema ments, as fibers
ute to the OP Hence, this not be neglec ncreasing the fect of softwa ng times great results call f nd protection ecovery impai icating the OL as software hinder this pro ssess for whic res become n y a feasibility/
REFE er et al., “An in
g., Vol. 45, Issue M. De Andrade, R s networks: archit l. 100, pp. 1188 – ost per client in sp
res heavily in ecially in den et in sparse sc X, and it can still account fo h the percentag
y due to softw arkably in den
s become mor PEX, albeit in work also sh cted when asse e duration of are failures. Y ter than 40000 for extending schemes. In irment must b LT is also seen failures tend otection schem ch failure inte negligible, y /economic ana
ERENCES ntroduction to PO
3, pp. S15 – S25, R. Roy and P. Ch tectures and capa – 1196, May 2012
parse scenarios for
ncrease the fa nse scenarios cenarios feede n be reduced for an importan
ge of business ware failures, i nse scenarios.
re prone to fa n a lesser way
hows that so essing the OP f the testing Yet this reduct
0 hours.
this work to Long-Reach P be further ana n as a cost-eff to propagate me. Also, it wo ensity (testing et this has alysis.
ON technologies, , Mar. 2007.
howdhury, “Evolu acity upgrades,” P
.
r different percen
failure- and if er fiber d with
nt part users.
it does Yet in ailures, y than
ftware PEX of phase tion is o other PONs, alyzed.
fective e, they ould be g time) to be
,” IEEE ution of Proc. of
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
ntages of business
J. Chen, L. W reliability perfo IEEE Commun.
E. Wong, “Surv multiplexed p Communications A. Fernandez an network geome Stockholm, Swe L. A. Barroso, Repairs,” in The of warehouse-sc 2013, ch. 3, sec.
A. Fernandez an Division Multip of RNDM, Barce A. C. Fadel, R embedded optic dos Campos, Br C. Jones and O ed., Addison We A. C. Fadel, R Software Valid Pasadena, USA, M. R Lyu (ed.), of Software Rel sect. 3.5.1, pp. 9 ITU-T Rec. G98 general characte B. E. Helvik, “F Communication Department of T G. J. Anders and for power system pp. 654 – 660, M D4.2.1: Techni optical access sy clients with vary
Wosinska, C. M.
ormance study o Mag., Vol. 48, Is vivable architectu passive optical
s, Vol. 325, pp. 1 nd N. Stol, “OPEX etric and Markov eden, 2014, pp. 25 J. Clidaras and U e datacenter as a
cale machines, 2n . 7.2.2, pp. 105 – nd N. Stol, “On th plexed Passive Op
elona, Spain, 201 R. Moraes and E cal network softw razil, 2011.
O. Bonsignour, Th esley Professiona R. Moraes, P. M dation of a GPO
, 2013, pp. 59.
, “Software Relia liability Engineer 98 – 99.
84.1, “Gigabit-cap eristics,” 2008, pp Fault tolerance,”
Networks Desig Telematics, NTNU d A. M. Leite da m equipment,” IE May 2000.
ical Assessment ystems concepts, ying testing time (
Machuca and f fiber access n ssue 2, pp. 56 – 65
ures for time an networks,”
52 – 159, Apr. 20 X simulation stud v cost models,”
52 - 257.
U. Hölzle, “Deal computer: an intr nd ed. Morgan &
107.
he Impact of Soft ptical Networks D 4, pp. 225 – 231.
E. Martins, “Au ware,” in Proc. of he Economics of al, 2011.
Martins and E.
ON Network,” i ability Modelling ring, 1st ed. Mcg pable passive opti p. 7.
in Dependable C gn and Evaluati U, 2009, ch. 2, sec
Silva, “Cost relat EEE Trans. Powe and compariso OASE Project, 20 (software failure i
M. Jaeger, “Co network architect
5, Feb. 2010.
d wavelength div Journal of O 014.
dy of PONs based in Proc. of ON ling with Failure roduction to the d
& Claypool Publi tware Failures on Dependability,” in utomated validati f LADC 2011, Sao f Software Qualit
Martins, “Autom in Proc. of ISS
Survey,” in Hand graw-Hill, 1996,
ical networks (GP Computing System ion, 5th ed., No ct. 2.3.2, pp. 97 – ted reliability mea er Syst., Vol. 15, N on of next-gener
011, pp. 37 - 38.
intensity).
st vs.
tures,”
vision Optics d on a NDM, es and design ishers, n Time Proc.
on of o Jose ty, 1st mating SREW, dbook ch. 3, PON):
ms and orway:
– 102.
asures No. 2, ration