Ci a ion: Gabilondo, Á.; Fe nández,
Z.; Viola, R.; Ma ín, Á.; Zo illa, M.;
Anguei a, P.; Mon albán, J. T a ic
Classi ica ion o Ne wo k Slicing in
Mobile Ne wo ks. Elec onics 2022,11,
1097. h ps://doi.o g/10.3390/
elec onics11071097
Academic Edi o s: Bin Han, Simon
Pie o Romano and Pa ick Seeling
Recei ed: 28 Janua y 2022
Accep ed: 29 Ma ch 2022
Published: 30 Ma ch 2022
Publishe ’s No e: MDPI s ays neu al
wi h ega d o ju isdic ional claims in
published maps and ins i u ional a il-
ia ions.
Copy igh : © 2022 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license (h ps://
c ea i ecommons.o g/licenses/by/
4.0/).
elec onics
A icle
T a ic Classi ica ion o Ne wo k Slicing in Mobile Ne wo ks
Ál a o Gabilondo 1,2,* , Zaloa Fe nández 1, Robe o Viola 1, Ángel Ma ín 1, Mikel Zo illa 1,
Pablo Anguei a 2and Jon Mon albán 3
1Vicom ech Founda ion, Basque Resea ch and Technology Alliance (BRTA), 20009 San Sebas ián, Spain;
[email p o ec ed] (Z.F.); [email p o ec ed] (R.V.); [email p o ec ed] (Á.M.);
[email p o ec ed] (M.Z.)
2Depa men o Communica ions Enginee ing, Uni e si y o he Basque Coun y (UPV/EHU),
48013 Bilbao, Spain; [email p o ec ed]
3Depa men o Elec onic Technology, Uni e si y o he Basque Coun y (UPV/EHU),
20018 San Sebas ián, Spain; [email p o ec ed]
*Co espondence: [email p o ec ed]g
Abs ac :
Ne wo k slicing is a p omising echnique used in he sma deli e y o a ic and can
sa is y he equi emen s o speci ic applica ions o sys ems based on he ea u es o he 5G ne wo k.
To his end, an app op ia e slice needs o be selec ed o each da a low o e icien ly ansmi da a o
di e en applica ions and he e ogeneous equi emen s. To apply he slicing pa adigm a he adio
segmen o a cellula ne wo k, his pape p esen s wo app oaches o dynamically classi ying he
a ic ypes o indi idual lows and ansmi ing hem h ough a speci ic slice wi h an associa ed 5G
quali y-o -se ice iden i ie (5QI). Finally, using a 5G s andalone (SA) expe imen al ne wo k solu ion,
we apply he adio esou ce sha ing con igu a ion o p io i ize a ic ha is dispa ched h ough
he mos sui able slice. The esul s demons a e ha he use o ne wo k slicing allows o highe
e iciency and eliabili y o he mos c i ical da a in e ms o packe loss o ji e .
Keywo ds: ne wo k slicing; 5QI; a ic classi ica ion; nDPI; 5G ne wo k
1. In oduc ion
Ne wo k slicing is a p omising echnique ha has he po en ial o accommoda e
indi idual a ics and sa is ies he quali y-o -se ice (QoS) equi emen s o he da a lows
o speci ic applica ions in a 5G ne wo k [
1
]. When applied o 5G ne wo ks, slicing can
be pe o med a di e en le els [
2
]: a he adio segmen , allowing o an e icien and
dynamic alloca ion o adio esou ces; in he mul i-access edge compu ing (MEC) esou ces,
when se ices a e pushed o he edge and need sma hos ing o balance pe o mance–cos
ade-o s; in he ne wo k co e, when a ic isola ion is equi ed o ensu e secu i y; o
du ing he deploymen o end- o-end solu ions, when mul i-domain/si e o mul i-use
endo pla o ms come in o play in p i a e ne wo ks.
In ecen yea s, in bo h he indus ial and au omo i e sec o s, he use o In e ne o
Things (IoT) connec ions has inc eased due o i s wide ange o applica ions. Mo eo e ,
his inc ease is expec ed o each a ound 50% o all de ices and connec ions wo ldwide
by 2023 [
3
]. Howe e , hese connec ions esul in cybe -physical scena ios whe e sen-
so ized da a a e u ned in o ac ionable da a, meaning ha asymme ic communica ions
will domina e and as amoun s o in o ma ion will be a ailable on uplinks (UL).
Fo hese sec o s, he 5G In as uc u e Public P i a e Pa ne ship (5G PPP) [
4
], and
he manu ac u ing (5G-ACIA [
5
]) and au omo i e (5GAA [
6
]) clus e s ha e p o ided all
he key pe o mance indica o s (KPIs) o ep esen a i e applica ions o each use case,
including enhanced mobile b oadband (eMBB), massi e machine- ype communica ions
(mMTC), and ul a- eliable low-la ency communica ions (URLLC) [7,8].
In his con ex , he nego ia ion and con ol messages and he ac ual da a lows may
ha e di e en communica ion needs, so he equi ed QoS g anula i y goes beyond he
Elec onics 2022,11, 1097. h ps://doi.o g/10.3390/elec onics11071097 h ps://www.mdpi.com/jou nal/elec onics
Elec onics 2022,11, 1097 2 o 27
senso de ice o he applica ion i sel . Addi ionally, he managemen o indi idual da a
lows becomes a majo conce n o Indus ial In e ne o Things (IIoT) and elema ic con ol
uni (TCU) a chi ec u es, in which wi ed o wi eless senso s and sys ems use ga eways
ha concen a e da a lows ha a e pushed o dis an sys ems o o he In e ne . Thus, a
single de ice could simul aneously p oduce o consume da a lows o di e en ypes wi h
speci ic QoS demands, beyond indi idual de ices o speci ic applica ions.
Fu he mo e, due o he a ie y o applica ions, p o ocols, and business logics in p i-
a e ne wo ks wi hin he indus ial and au omo i e sec o , he abili y o classi y indi idual
da a lows is essen ial. The e o e, ins ead o mapping applica ion lows o indi idual
de ices, g anula i y can be used o classi y indi idual da a lows in hese en i onmen s.
In his way, i becomes logical o assign a highe p io i y o con ol messages o e o he
da a lows whe e he s a is ics o he a ic may p o ide a clea iew o he c i icali y o
sho messages sen om ime o ime (likely URLLC), on he bandwid h equi ed o da a
lows sending high olumes o packe s all he ime (likely eMBB) and o he high densi y
o messages om mul iple pa ies sending equen sho messages (likely mMTC).
Due o he asymme y o senso ized communica ions as well as he exis ence o lows
wi h di e en QoS equi emen s, he use o adio access ne wo k (RAN) slicing may be
bene icial [9].
The objec i e is o empowe he RAN o become awa e o he a ic ea u es o
indi idual da a lows, enabling hei classi ica ion and hei dynamic assignmen o exis ing
slices. To his end, he majo con ibu ions o ou wo k include wo o iginal app oaches o
handling he di e en ypes o a ics when applying he ne wo k slicing echnique in UL.
These con ibu ions can be summa ized as ollows:
•
On he one hand, a mechanism based on a a ic classi ie is used o iden i y he class
o indi idual da a lows. This allows o subsequen slicing o be dynamic, adap i e
and anspa en o end use s by being able o ansmi all ypes o a ic.
•
On he o he hand, he p oposed app oaches assign, in eal ime, he da a low o
one o he a ailable slices associa ed wi h a 5G QoS iden i ie (5QI) alue om [
10
],
acco ding o speci ic policies. He e, he da a low classes conside ed a e he main ones
p esen in he indus ial and au omo i e sec o s, including con ol messaging, ideo
s eaming, IoT da a, and gene ic web da a.
The con ibu ion o his pape goes beyond he simple use o a ic classi ica ion and
ne wo k slicing echniques. A dynamic and anspa en alloca ion o slices a he adio
le el was ca ied ou in a eal 5G ne wo k deploymen . This eal deploymen also ma ks
a di e ence om o he p oposed s a e-o - he-a solu ions, which emain a he le el o
simula ion. In o de o e alua e he pe o mance o he p oposed solu ions, a scena io was
de ined in which UL communica ions a e sa u a ed in a eal es bed o a 5G s andalone
(SA) ne wo k. The esul s we e also compa ed wi h a e e ence scena io whe e no slicing
was applied.
The es o his pape is o ganized as ollows. Fi s , ela ed wo k is p esen ed in
Sec ion 2, co e ing he concep o ne wo k slicing, he a ic ypes, and he me hods o
classi ica ion. In Sec ion 3, he mos ele an so wa e packages o a 5G expe imen al ne -
wo k wi h ne wo k slicing suppo a e desc ibed. Then, Sec ion 4p esen s he RAN slicing
solu ion and he algo i hms in ol ed in i s o mula ion. Fu he mo e, he e alua ion se up,
he sys em pe o mance analysis, he de ined me ics, and he esul s o he expe imen
ca ied ou a e desc ibed in Sec ion 5. Finally, he esul s a e discussed in Sec ion 6, and
Sec ion 7concludes he pape .
2. Backg ound and Rela ed Wo k
2.1. Ne wo k Slicing
Ne wo k slicing is a echnique in oduced by he NGMN (Nex Gene a ion Mobile
Ne wo k) in [
11
] and allows o mul iple i ual ne wo ks o un in a single common
physical in as uc u e in an e icien and cos -e ec i e manne , while sa is ying di e en
se s o QoS. A ne wo k slice is de ined as a se o a ailable esou ces (ne wo k, compu a-
Elec onics 2022,11, 1097 3 o 27
ion, and adio) assigned o a i ualized ne wo k se ice o sa is y speci ic equi emen s
associa ed wi h he se ice. In u n, ne wo k slicing consis s o de ining a se o policies and
mechanisms o iden i y he a ic lows associa ed wi h each slice. Taking his in o accoun ,
a common physical in as uc u e p o ides access o mul iple slices simul aneously in o de
o sa is y he QoS o de ices, use s, o applica ions, as shown in Figu e 1.
Figu e 1. 5G ne wo k slicing a chi ec u e.
Cu en ly, each use equipmen (UE) elemen can be se ed by eigh slices, and he
iden i ie o each slice is called he Single Ne wo k Slice Selec ion Assis ance In o ma ion
(S-NSSAI), whe e an NSSAI is a collec ion o hese slices. This iden i ie is employed by he
5G Co e (5GC), he 5G-RAN, and he UE elemen s o a 5G ne wo k.
In he same way, each iden i ie consis s o wo pa s. Fi s , he slice/se ice ype (SST)
iden i ies he slice in e ms o i s cha ac e is ics and se ices. Second, he slice di e en ia o
(SD) allows o di e en ia ing mul iple slices om he same SST. While he SST pa ame e
is manda o y, he SD is op ional and is used o di e en ia e be ween wo slices wi h he
same SST. Mo eo e , some SST s anda d alues ha e been se by he 5G use cases, as
Table 1shows.
Table 1. S anda dized SST alues [10].
SST Slice/Se ice Type
1 eMBB
2 URLLC
3 Massi e IoT (MIoT)
4 Vehicle o e e y hing (V2X)
5 High-Pe o mance Machine-Type Communica ions (HMTC)
The 5G end- o-end ne wo k slicing equi es his concep o be applied in he adio,
edge, co e, and anspo ne wo ks.
F om he co e ne wo k pe spec i e, ne wo k slicing mainly p o ides he possibili y o
deploying mul iple ins ances o i ual 5GCs concu en ly on a single common physical
in as uc u e [
12
]. Each o hese ins ances is con igu ed o sa is y di e en se ice equi e-
men s. In 5G ne wo ks, he design o he co e ne wo ks is implemen ed as i ualized
ne wo k unc ions (VNF), ollowing he ne wo k unc ion i ualiza ion (NFV)/so wa e-
de ined ne wo k (SDN) pa adigm [
13
]. Mo eo e , NFV and SDN a e he pilla s ha enable
ne wo k slicing in 5G ne wo ks and c ea e slicing in a anspo ne wo k segmen o mee
he equi emen s and he QoS equi emen s o he applica ion and se ices [14].
Elec onics 2022,11, 1097 4 o 27
F om he RAN poin o iew, he ne wo k slicing concep is implemen ed conside ing
dynamic esou ce managemen mechanisms. These mechanisms allow o an e icien and
dynamic alloca ion o adio esou ces, i.e., h ough sma schedule s a he media access
con ol (MAC) le el. To ma ch he equi emen s, his esou ce alloca ion mus conside he
KPIs o he di e en se ices/applica ions ha a e se ed by he slice.
In o de o exempli y he applica ion o slices in he p e iously de ined ne wo k
segmen s on he 5GC, he edge, and he anspo ne wo k segmen s, he CPU esou ces,
he ne wo k opology used, o he a ic cha ac e is ics can be conside ed, i.e., each slice can
alloca e CPU esou ces o ob ain i s opology. Mo eo e , each a ic ype can be associa ed
wi h a slice isola ed om he es o he ne wo k [
15
]. In con as , on he 5G RAN segmen ,
he a ailable bandwid h esou ces can be assigned e icien ly among di e en slices while
conside ing hei equi emen s.
Fu he mo e, in e ms o esou ces, he e a e h ee a ailable op ions conce ning he
slice isola ion: (i) ully sha ed esou ces, (ii) pa ially sha ed esou ces, and (iii) comple ely
dedica ed esou ces [
12
,
16
]. He e, di e en slices a e isola ed as long as ac ions pe o med
on one slice do no a ec he pe o mance o ano he .
In he case o comple ely dedica ed esou ces, each slice has a se o adio esou ces
assigned in he con ol and use plane as well as in he MAC-le el schedule and in he
adio spec um. In his case, each slice has access o a pe cen age o dedica ed physical
esou ce blocks (PRBs), which a e de ined as minimum uni s o esou ces ha a base s a ion
can alloca e o a UE, gua an eeing bo h a ic isola ion be ween slices and ha he QoS
equi emen s (e.g., delay and capaci y cons ain s) a e me . Howe e , his educes lexibili y
as i will no allow esou ces o be mo ed om one slice o ano he , and he esou ces will
be was ed i a slice is no used.
In his pape , we ocus on he adop ion o ully sha ed esou ces o ne wo k slicing
in he RAN segmen . In he ully sha ed esou ce op ion, all he slices sha e he adio
spec um, he MAC-le el schedule , and he 5G ne wo k con ol plane. The PRBs a e
managed by a common schedule , which alloca es esou ces o each slice acco ding o he
equi emen s and QoS equi ed by hei se ices, applica ions, o use cases. I should be
no ed ha nei he he a ic isola ion no he a ge QoS le els a e ensu ed in he ully
sha ed esou ces op ion. Howe e , since he esou ces a e sha ed among all slices, i is
a mo e lexible op ion, since i allows PRBs o be dynamically alloca ed acco ding o he
needs o he applica ion.
Wi h his in mind, se e al EU ini ia i es ha e ocused on de ining 5G ne wo k slicing
a chi ec u es. Some examples include 5G NORMA [
17
], SliceNe [
18
], and 5GTANGO [
19
].
In he case o 5G NORMA, a mul i- enan and mul i-se ice 5G sys em a chi ec u e based
on he concep o ne wo k slicing is p oposed [
20
]. SliceNe aims o achie e end- o-end
ne wo k slicing h ough con ol, managemen , and o ches a ion mechanisms [
21
]. In he
case o 5GTANGO, a ne wo k slicing esou ce alloca ion and moni o ing amewo k o e
mul iple clouds and ne wo ks was c ea ed [
22
]. In con as o he ocus o his pape , which
is RAN slicing, all o hese ini ia i es ocused mo e on he applica ion o ne wo k slicing
using SDN and NFV, igno ing he adio pa adigm.
Se e al pape s ha e conside ed he challenges and esea ch issues aised due o
he applica ion o he ne wo k slicing concep in a 5G ne wo k [
1
,
12
,
13
,
16
,
23
–
26
]. Ea ly
con ibu ions ela ed o he concep o ne wo k slicing we e ocused on LTE [
27
], bu his
end has shi ed in ecen yea s o 5G communica ions. Howe e , i mus be unde lined
ha cu en ne wo k slicing in 5G ne wo ks has mainly been ca ied ou and e alua ed in
he con ex o he 5GC segmen , hus neglec ing he adio segmen , which is he ocus o his
pape [
28
]. Fu he mo e, many o he con ibu ions a he RAN le el ocus on heo e ical
analyses o simula ion-le el e alua ions [
29
–
32
]. Thus, he e alua ion o he solu ions is
no pe o med in a eal 5G a chi ec u e.
One o he open issues is he lack o analysis adap ed o ypes o da a lows in o de
o de ine he esou ces equi ed in each slice, acco ding o he cha ac e is ics o he ype
o a ic o be ansmi ed. Fo his eason, he ocus o his pape is he map o da a low
Elec onics 2022,11, 1097 5 o 27
ypes o speci ic 5QI slices and i s e alua ion in a eal 5G ne wo k wi h ne wo k slicing
echniques a he RAN le el. I should be no ed ha he solu ion p esen ed in his pape is
also in line wi h he co e challenges desc ibed in [33].
2.2. T a ic Types
This sec ion desc ibes he ypes o a ic widely conside ed in he li e a u e; he
cha ac e is ics a e summa ized in Table 2[34] and can be desc ibed as ollows:
•
Video a ic is he s eaming o ideo da a be ween endpoin s. This ype o a ic is
loss- ole an bu sensi i e o delays in eal- ime s eaming.
•
Audio/ oice a ic is he ansmission o audio da a be ween endpoin s. Simila
o ideo a ic, his ype o a ic is loss- ole an bu sensi i e o delays in eal-
ime s eaming.
•
IoT a ic is he s eaming o cybe -physical da a and in o ma ion abou he en i-
onmen collec ed by hund eds o housands o de ices and ansmi ed pe iodically.
This kind o a ic is no e y loss- ole an . Howe e , ensu ing a high numbe o
connec ions while gua an eeing bandwid h and la ency is e y impo an .
•
WebDa a a ic is he amoun o da a sen and ecei ed by isi o s o a websi e. The
ypical size o he da a is a iable, as i mainly depends on he ype o websi e om
which he da a o igina es. By con as , he main ocus o his a ic is on bandwid h.
•
Da a ans e a ic e e s o he ansmission o iles o a ying sizes be ween end-
poin s. Fo his ype o a ic, he mos impo an equi emen is he eliabili y o he
con en upon ecep ion, wi h longe delays being mo e ole able han o o he ypes
o a ic.
•
Con ol da a a ic con ains ne wo k con ol messages. These messages a e usually
p oduced in low olumes wi h e y low packe sizes bu ha e s ong deli e y equi e-
men s, o example, a e y long delay o he ne wo k con ol ames expe iencing a
loss can lead o loss o ne wo k unc ions.
• Bes -e o a ic comp ises a ic ha has no speci ic equi emen s o be ul illed.
Table 2. T a ic- ype cha ac e is ics [34].
T a ic Cha ac e is ics
T a ic Tole ance
o Loss
Ne wo k
Requi emen
Gua an ee
Packe Size
(By es)C i icali y Pe iodici y
Video Yes La ency High
(1000∼1500) Medium Pe iodic
Audio/Voice Yes La ency Va iable
(20∼1500) Medium Pe iodic
IoT No Nº o
connec ions
Low
(50∼500) Low Pe iodic
WebDa a No Bandwid h Va iable
(30∼1500) Low Spo adic
Da a T ans e No Bandwid h Va iable
(30∼1500) Low Spo adic
Con ol Da a No La ency Ul a low
(30∼150) High Pe iodic
Bes -E o Yes None Va iable
(30∼1500) None Spo adic
2.3. T a ic Classi ica ion
This sec ion desc ibes he au oma ic a ic classi ica ion p ocess, he di e en g oups
in o which he algo i hms can be di ided, and he se o open-sou ce al e na i es conside ed
in his pape . Choosing he p ope a ic classi ica ion algo i hm is indispensable o
ob aining a p ope and e icien a ic classi ica ion.
Elec onics 2022,11, 1097 6 o 27
In gene al, he a ic classi ica ion p ocess ca ied ou by hese algo i hms in ol es
ou s eps. Fi s , he a ic coming om he ne wo k is collec ed o o m a da ase . The
second s ep ex ac s and selec s pa icula ea u es om he da ase ob ained in he p e ious
s ep. The hi d s ep akes in o accoun he ea u es ob ained in he p e ious s ep o iden i y
he a ic ca ego y. To do so, i uses pa e ns o model aining in conjunc ion wi h machine
lea ning algo i hms. Finally, in he las s ep, he ob ained esul s a e e i ied.
Taking his in o accoun , he algo i hms can be di ided in o i e main g oups [35]:
•
S a is ics-based classi ica ion: his uses s a is ical in o ma ion ela ed o he a ic
wi hou analyzing he packe payload. Algo i hms belonging o his g oup ha e a high
compu a ional o e head as an analysis o heu is ics o indi idual packe s
is equi ed.
•
Co ela ion-based classi ica ion: his uses he co ela ion o lows combined wi h he
s a is ical in o ma ion o he a ic. As in he case o he s a is ics-based classi ica ion
g oup, hese also ha e a high compu a ional o e head.
•
Beha io -based classi ica ion: his uses hos in e ac ion and connec ion da a o classi y
a ic. Despi e he good esul s in e ms o accu acy and he ligh weigh p ocessing
demand, he a ic classi ica ion esul does no p o ide much de ail.
•
Payload-based classi ica ion: his uses he con en o he payload o some speci ic ields
o he payload o pe o m a ic classi ica ion. The e o e, he accu acy ob ained by
hese algo i hms is he highes among he g oups men ioned in his pape . This g oup
can be u he di ided in o deep packe inspec ion (DPI) algo i hms and s ochas ic
packe inspec ion (SPI) algo i hms. The di e ence be ween hem is he way in which
hey inspec he con en o he packe . While DPI algo i hms a e gene ally no able o
classi y enc yp ed packe s, SPI algo i hms can deal wi h enc yp ed a ic. Bo h ha e
high compu a ional o e heads.
•
Po -based classi ica ion: his uses only he po o classi y he a ic. This me hod
is he leas accu a e among all o he abo e g oups since many a ic sou ces use
dynamic po s o wo a ic sou ces may use he same po o ansmi da a wi h
di e en p o ocols. The e o e, in hese cases, he esul s o he classi ica ion may no
be co ec .
Conside ing he a o emen ioned ad an ages and disad an ages o he algo i hms
lis ed, DPI is he mos accu a e classi ica ion algo i hm, wi h an accep able compu a ional
bu den. The e o e, his pape ocuses on DPI algo i hms. Many p oduc s, bo h comme cial
and open-sou ce, ely on some o m o DPI o pe o m he classi ica ion p ocess. This
pape emphasizes he use o open-sou ce al e na i es because o he lexibili y and low
cos o e ed du ing implemen a ion. Mo e p ecisely, he pape ocuses on i e open-sou ce
ools: L7- il e , OpenDPI, Libp o oiden , nDPI, and NFS eam.
L7- il e [
36
]: his classi ie iden i ies packe s based on applica ion laye da a o de-
e mine which p o ocols a e being used, hus complemen ing classi ie s ha ma ch IP
add esses and po numbe s. This classi ie is mo e demanding in e ms o p ocessing and
memo y han o he s, so i s use is ecommended only when i is necessa y o map p o ocols
ha use unp edic able po s, o ma ch a ic on non-s anda d po s, o o dis inguish
be ween p o ocols ha sha e a po . Howe e , he p ojec has now been discon inued and
is conside ed closed.
OpenDPI [
37
]: his is an open-sou ce e sion o an ea ly e sion o PACE, a comme cial
classi ie , which has been s ipped o enc yp ed p o ocols, making i a mo e limi ed classi ie
wi h ewe suppo ed p o ocols and slowe pe o mance. As in he p e ious case, his
p ojec is closed and conside ed obsole e.
Libp o oiden [
38
]: his classi ie , simila o he o he s men ioned in his sec ion,
pe o ms payload-based classi ica ion, bu wi h one ele an di e ence. This ool examines
only he i s ou by es o he payload in each di ec ion. This minimizes he s o age
capaci y needed o s o e he packe aces, as well as he compu a ional load equi ed
o classi ica ion. Payload pa e n ma ching, payload size, po numbe s, and IP add ess
ma ching a e used when pe o ming he classi ica ion.
Elec onics 2022,11, 1097 7 o 27
nDPI [
39
]: his is an ex ension o OpenDPI, wi h u he op imiza ion o bo h pe o -
mance and speed, as well as a la ge numbe o p o ocols, including enc yp ed p o ocols
wi h he addi ion o a decode o a secu e socke s laye (SSL). I is based on a a ic classi i-
ca ion lib a y using bo h packe heade and payload. Mo eo e , i allows o decoding open
sys em in e connec ion (OSI) laye s 3 and 4 o he packe s o imp o e he classi ica ion. I
is a p ojec wi h cons an upda es and a la ge communi y. In addi ion, he ou pu o he
esul s is p esen ed in a simple o m o u he p ocessing.
NFS eam [
40
]: his is based on he nDPI classi ie , al hough i is based on Py hon,
unlike p e ious ools ha a e based on C o C++. This nDPI-based classi ica ion allows
NFS eam o pe o m a eliable iden i ica ion o enc yp ed applica ions and me ada a
inge p in ing. In addi ion, his classi ie is lexible and allows o he c ea ion o new
ea u es by adding lines and Py hon code in a simple way.
A compa ison o he mul iple classi ie s p esen ed is shown in Table 3.
Table 3. Compa ison be ween di e en DPI algo i hms [41].
Name Released Upda es Language Apps/P o ocols
Iden i ied
L7- il e 2009 Dep eca ed C++ ∼110
OpenDPI 2011 Dep eca ed C ∼100
Libp o oiden 2013 2–3 qua e C++ ∼250
nDPI 2014 Few days C ∼170
NFS eam 2019 Few days Phy on ∼180
Al hough he OpenDPI and L7- il e p ojec s we e abandoned and hei de elopmen
s opped se e al yea s ago, hey ha e been included in his pape o e e ence because
many scien i ic pape s base hei esul s and conclusions on hese wo classi ie s.
Among all he op ions men ioned abo e, he open-sou ce classi ie s ha dig deep in o
he packe s o classi ica ion a e nDPI and NFS eam. Bo h o hem allow a la ge numbe o
p o ocols o be classi ied, including hose necessa y o he ypes o a ic s udied in he
p e ious sec ion, such as Real Time P o ocol (RTP) o Video, Message Queue Teleme y
T anspo (MQTT) o IoT, and Hype ex T ans e P o ocol (HTTP) o WebDa a. They also
ha e a la ge communi y ha o e s suppo when p oblems a ise and a e upda ed om ime
o ime. Howe e , al hough NFS eam is somewha mo e lexible in e ms o da a ou pu ,
he nDPI classi ie displays mo e de ailed in o ma ion and allows o be e p ocessing.
3. Ne wo k Slicing Solu ions
In his sec ion, we b ie ly desc ibe he mos ele an so wa e o deploying 5G SA
expe imen al ne wo ks in which a ne wo k slicing deploymen can be pe o med. Table 4
shows he analysis o solu ions o es beds.
Elec onics 2022,11, 1097 8 o 27
Table 4. Analysis o ne wo k slicing solu ions.
Name Desc ip ion P os Cons
Open5GCo e
[42]
Comme cial
so wa e o
5GC [43]
Allows i ualiza ion o
he so wa e
The same le el o bandwid h,
h oughpu , la ency, e c., o
each slice. So, no di e ence be-
ween slices.
License needed
Open5Gs [
44
]
Open-sou ce
so wa e o
5GC wi h C
language
Allows o i ualiza ion
o he so wa e.
The numbe o UEs pe
slice is no limi ed
Only 5GC deploymen wi h-
ou RAN
Ama iso
[45]
Comme cial
p oduc s o
5GC, RAN,
and UEs
Suppo s i ualiza ion
depending on he license
pu chased.
Allows end- o-end 5G
ne wo k deploymen
wi h ne wo k slicing
Radio ne wo k slicing is ap-
plied in he schedule o he
base s a ion called gNodeB
(gNB) h ough a ic p io i i-
za ion
O he ele an open-sou ce solu ions o academic and esea ch communi ies include
5G OpenAi In e ace (OAI) [
46
] and ee5GC [
47
]. Ne e heless, al hough hey allow o
he deploymen o 5G SA ne wo ks—end o end in he case o 5G OAI and only he co e
sys em in he case o ee5GC—in con as o he o he so wa e solu ions men ioned abo e,
hey canno cu en ly implemen unc ional ne wo k slicing.
Conside ing he p e ious analysis, he comme cial so wa e Ama iso has been se-
lec ed as i suppo s ne wo k slicing based on a ic p io i iza ion a he RAN le el and
o e s a 5GC, a RAN, and UEs—all wi h ne wo k slicing suppo —unlike he o he solu ions
men ioned abo e.
Fu he mo e, he deploymen o he ne wo k slicing o e ed by Ama iso includes
a mechanism a he gNB schedule ha p io i izes a ic—bo h UL and downlink (DL)—
based on he 5QI associa ed wi h each slice. The 5QI is an indica o o a se o QoS cha ac e -
is ics such as he p io i y le el, packe delay, packe e o a e, e c. These QoS cha ac e is ics
can ei he be s anda dized o non-s anda dized. The s anda dized 5QI alues can be seen
in [
10
], while he non-s anda dized 5QI can be used o ad hoc con igu a ions. Mo eo e ,
he lowe he p io i y alue assigned by he 5QI, he highe he p io i y o he slice.
Fo he applica ion o ne wo k slicing in he RAN segmen , no only should he RAN
be awa e o he cha ac e is ics and pa ame e s o each o he slices bu he UE mus also
know hem. The e o e, he UE needs o disco e bo h he iden i ie s o he slices (SST and
SD) and he 5QI pa ame e o each o hem, which will be used o he p io i iza ion o
di e en slices. Fo his, he sequence o messages o disco e ing hese pa ame e s is
as ollows:
1.
The con ex is equi ed o ind ou which slices a e in he 5GC da abase o ha speci ic
UE. These slices, as shown in Figu e 2, a e iden i ied wi h hei co esponding SST and
SD using he NG Applica ion P o ocol (NGAP) Ini ial Con ex Se up Reques message.
2.
The connec ion o a slice is es ablished, as in Figu e 3—in his example, a slice wi h
an SST 7 and an associa ed 5QI o 70—using a message PDU Session Resou ce Se up
Reques , whe e he UE ecei es he 5QI pa ame e o ha slice. The e o e, o compile
all o he 5QI pa ame e s o he a ailable slices, i would be necessa y o p e iously
connec o each o hem.
Elec onics 2022,11, 1097 9 o 27
Figu e 2. Ini ial Con ex Se up Reques message wi h he allowed SST.
Figu e 3. PDU Session Resou ce Se up Reques message wi h he selec ed SST and associa ed 5QI.
Fo mo e in o ma ion on he message chain o making connec ions in 5G ne wo ks,
please e e o he echnical epo ound in [48].
4. P oposed App oaches o Ne wo k Slicing
4.1. Ta ge Scena io
The app oach p oposed o he applica ion o ne wo k slicing in his pape is explained
below. The app oach p esen ed he e uses ne wo k slicing in a scena io whe e a UE is
Elec onics 2022,11, 1097 16 o 27
Table 6. Bandwid h ansmi ed in he sa u a ion es .
Time (s) Bandwid h (Mbps)
Slice 1 Slice 2 Slice 3 Slice 4
0–20 20 20 20 20
20–40 40 20 20 20
40–60 60 20 20 20
60–80 80 20 20 20
80–100 100 20 20 20
100–120 20 20 20 20
120–140 20 40 20 20
140–160 20 60 20 20
160–180 20 80 20 20
180–200 20 20 20 20
200–220 20 20 40 20
220–240 20 20 60 20
240–260 20 20 20 20
260–280 20 20 20 40
280–300 20 20 20 20
This es was pe o med en imes o minimize any andomness and was ca ied ou
in he h ee men ioned scena ios. The numbe o epe i ions was selec ed a e obse ing
ha , e en wi h a la ge numbe o es s, he a iabili y was no app eciable. The e o e, we
concluded ha hese en es s we e su icien o ob ain eal and accu a e esul s in each
scena io. In he i s scena io, which wo ks as a e e ence, he ne wo k slicing echnique
was no used; he e o e, ou di e en UEs we e connec ed o he RAN and conges ed
one by one, acco ding o he me hodology al eady explained. In he second scena io, he
same ou UEs we e connec ed o he RAN. He e, each one was associa ed wi h a di e en
slice, which was conges ed depending on he p io i iza ion o he slices. Finally, in he las
scena io, one UE was connec ed o he same ou slices o he p e ious scena io, conges ed
in he same manne as in he second scena io.
5.2. Expe imen al Resul s
In his sec ion, he pe o mance o he wo p oposed app oaches desc ibed in
Sec ion 4
in e ms o bandwid h, packe loss, and ji e a e e alua ed using he sa u a ion es
desc ibed in he UL. As men ioned, o compa e he esul s, a e e ence scena io in which
ne wo k slicing was no conside ed was also e alua ed. All e alua ions we e pe o med
on a 5G SA ne wo k, as s a ed p e iously.
5.2.1. Bandwid h
Figu es 6–8show he bandwid h ob ained e e y second o each slice and each scena io.
Figu e 6shows he bandwid h measu ed by ou equal UEs in a scena io whe e he ne wo k
slicing concep was no applied, i.e., whe e all use s and a ic had he same p io i y. As
can be seen, he pe o mance o each use was he same; all use s sha e he bandwid h and
when a use , ega dless o who hey a e, asks o mo e, hey can use he ex a bandwid h
a ailable. In he same way, i a sa u a ion si ua ion is eached, he es o he use s a e no
a ec ed, as seen h oughou Figu e 6, in which 20 Mbps a e gua an eed in each slice.
In his i s scena io, al hough he o al bandwid h sha ed by all he UEs should be
a ound 105 Mbps, as men ioned abo e, i can be seen ha when a UE ansmi s mo e
han he ini ial 20 Mbps ansmi ed pe slice and, a some poin s o he es , exceeds
ha heo e ical limi , only 6–7 Mbps can be ob ained in addi ion o he ini ial 20 Mbps
ansmi ed. This implies ha he o al bandwid h o he ne wo k is a ound 85–90 Mbps.
The eason o his beha io is ha , when sha ing bandwid h be ween mul iple use s,
conside ing how scheduling wo ks in 3 d Gene a ion Pa ne ship P ojec (3GPP) sys ems,
i is no as simple as di iding he adio esou ces equally. A he same ime, i is necessa y
o ake in o accoun ha he es s we e ca ied ou in a wi eless en i onmen using an
Elec onics 2022,11, 1097 17 o 27
expe imen al 5G ne wo k deploymen , whe e he e is mo e ins abili y o conside han in a
wi ed en i onmen .
Figu e 6. Bandwid h ob ained om scena io wi hou p io i iza ion.
Figu e 7. Bandwid h ob ained om he scena io using he i s app oach.
Figu e 7also ep esen s he bandwid h measu ed in he same ou UEs bu in a
scena io whe e he ne wo k slicing concep was applied. Mo e speci ically, each UE is
associa ed wi h a di e en slice. In his igu e, i can be seen ha he pe o mance o each
use is di e en . The use wi h he highes p io i y, associa ed wi h he Con ol slice, has
he expec ed pe o mance and can eques mo e bandwid h by aking bandwid h away
om he o he UEs/slices. When he use wi h he second-highes p io i y, connec ed o
he S eaming slice, eques s 40 and 60 Mbps ( om 2:00 o 2:20 and om 2:20 o
2:40 min
,
espec i ely), hey ake hem om he hi d and ou h use s, connec ed o he IoT and
WebDa a slices, espec i ely, wi hou a ec ing he highes p io i y slice. Howe e , when
hey ask o 80 Mbps ( om 2:40 o 3:00 min), hey canno ob ain mo e bandwid h as
he a ic has become conges ed. Rega ding he hi d slice, when a use asks o mo e
bandwid h, hey ake away pa o he lowes p io i y slice, which is only abou 30 Mbps in
he bes case. Fo ou h slice, when a use asks o mo e bandwid h, hey canno ake any
om he i s , second, o hi d slices since hese ha e highe p io i ies.
This beha io , whe e he second slice canno ask o mo e han 80 Mbps and he
hi d and ou h slices canno ask o mo e bandwid h, p o ides he same esponse as
he p e ious scena io, since ha ing o sha e he bandwid h among se e al UEs makes i
impossible o achie e he ull nominal bandwid h o 105 Mbps. The only case whe e a slice
comes close o eaching his sco e is in he highes p io i y slice, as his can ake bandwid h
om he o he s. Howe e , e en his si ua ion has some ins abili ies when ansmi ing
100 Mbps and canno main ain his le el o ansmission.
Elec onics 2022,11, 1097 18 o 27
Finally, Figu e 8shows he bandwid h ob ained in he hi d scena io, whe e a single
UE is connec ed o he same ou slices used in he second scena io, i.e., Con ol, S eaming,
IoT and WebDa a. This igu e clea ly shows he beha io o a ic p io i iza ion in he
di e en slices. I can be seen ha , when each slice equi es mo e bandwid h, i ob ains he
bandwid h om lowe p io i y slices. Thus, he Con ol slice eaches he sa u a ion limi s
as he channel in 100 Mbps, once i ob ains bandwid h om o he slices. The S eaming
slice ob ains i s bandwid h om he IoT and WebDa a slices, a a maximum o 80 Mbps,
bu i does no a ec he bandwid h o he Con ol slice. The IoT slice ob ains i s bandwid h
om he WebDa a slice, wi h a maximum o 60 Mbps, wi hou a ec ing he Con ol and
S eaming slices. Finally, he WebDa a slice is he mos uns able because i can only access
a ailable bandwid h om he h oughpu , no ha in use. The e o e, i can only ob ain
40 Mbps a mos .
As can be seen, in his case, a bandwid h o mo e han 100 Mbps is ob ained, since,
unlike he o he scena ios, his bandwid h is dedica ed o a single UE, ac ing as a ga eway
o he di e en da a lows dis ibu ed among he di e en slices.
Figu e 8. Bandwid h ob ained om he scena io using he second app oach.
5.2.2. Packe Loss
The igu es shown in his subsec ion (Figu es 9–11) show he pe cen age o packe s los
om each slice in each o he scena ios. Howe e , a some speci ic imes, when he di e en
slices a e se e ely conges ed, long delays occu and he ecep ion is e a ic when measu ing
packe loss; he e o e, ewe packe s a e measu ed as sen . This makes he esul s di icul
o in e p e , mainly in he lowe p io i y slices such as IoT and WebDa a. Due o his, and
in o de o ob ain a mo e uni o m ep esen a ion o compa ison, smoo hing was ca ied
ou by ob aining and modeling he a e age alue o he esul s o each slice e e y 5 s.
Figu e 9p esen s he packe losses in he i s scena io wi hou applying he ne wo k
slicing concep . In his igu e, i can be seen ha he losses inc ease o each use each ime
ha he use eques s mo e bandwid h. This is consis en wi h he esul s shown in
Figu e 6.
As men ioned abo e, in his scena io, each use canno ob ain mo e han
26–27 Mbps
in
o al, so when a use eques s 40 Mbps, he losses o ha use a e a ound 30% and he
o he s a e no a ec ed. When one o he use s asks o 60 Mbps, he losses a e a ound 55%;
when he demand inc eases o 80 Mbps, he losses a e a ound 66%, and when he use asks
o 100 Mbps, he losses a e a ound 75%. This beha io is he same o all use s.
Elec onics 2022,11, 1097 19 o 27
Figu e 9. Packe loss ob ained om he scena io wi hou p io i iza ion.
Figu e 10 shows he packe loss in he second scena io, whe e each use is connec ed
o a di e en slice. Fi s , ega ding he ep esen a ion o IoT and WebDa a, we obse e
ha he e a e in e als (1:00–1:40 o IoT and 2:40–03:00 o WebDa a) in which, al hough
he slices a e sa u a ed, he packe losses a e lowe han hose o o he highe p io i y
slices, such as o he S eaming slice in he i s in e al o o he IoT slice in he second
in e al. This is because, al hough he losses indica e a lowe pe cen age, he o al numbe
o packe s ecei ed in he slices du ing hose in e als is also lowe , implying a highe loss.
An example o his beha io is a 1:20 in he S eaming slice. In his case, he e is a loss
pe cen age o 53% o e a o al o 1506 da ag ams ecei ed, while in he IoT slice, he e is
a loss pe cen age o 33% o e a o al o 66 da ag ams ecei ed. This shows ha he o al
sa u a ion o he highe p io i y slice means ha many o he packe s ansmi ed by he
lowe p io i y slices do no e en each he Ipe se e , hus inc easing packe losses. I is
wo h emembe ing ha bo h slices y o send 20 Mbps du ing his sa u a ion condi ion in
his ime in e al.
Taking his in o accoun when analyzing his case, i can be seen ha he losses a e
also consis en wi h he beha io shown in Figu e 7, which shows he esul s ela ed o
bandwid h in his second scena io. I can be seen ha , as a use becomes sa u a ed, he
o he lowe p io i y use s lose mo e packe s, s a ing wi h he lowes p io i y use . I is
also wo h no ing he pe cen ages o packe loss o he Con ol slice a 1:20–1:40, o he
S eaming slice a 2:40–3:00, o he IoT slice a 3:20–4:00, and o he WebDa a slice a
4:20–4:40 a e due o he a o emen ioned sha ing o a ailable bandwid h be ween mo e han
one use .
Figu e 10. Packe loss ob ained om he scena io using he i s app oach.
Finally, Figu e 11 shows he packe losses in he hi d scena io, whe e one use is
connec ed o ou slices. In his las scena io, i can be obse ed ha when a slice, ega dless
Elec onics 2022,11, 1097 20 o 27
o i s p io i y, demands mo e bandwid h, he packe losses a ec only he slices ha ha e
a lowe p io i y. The e o e, in his case, unlike he o he scena ios, he slice asking o
mo e bandwid h ne e loses packe s. This allows he highes p io i y slice o achie e high
s abili y and eliabili y by no being sa u a ed a any poin in he es . Addi ionally, as in
he p e ious scena io, a a highe sa u a ion le el, he lowe p io i y slices de ec ewe
packe s and pe cei e ewe packe s as being los . As an example, a 1:30, he S eaming
slice has 77% losses ou o 1923 da ag ams de ec ed, while IoT and WebDa a ha e 72% and
52% losses ou o 136 and 33 da ag ams de ec ed, espec i ely.
Figu e 11. Packe loss ob ained om he scena io using he second app oach.
5.2.3. Ji e
In his las pa o he esul s, Figu es 12 and 13, as well as Tables 7and 8show
compa isons be ween he h ee scena ios conside ed. The compa isons ocus on he ji e
ob ained o bo h he Con ol and S eaming slices. The ji e ob ained e e s o he a ia ion
in delays in a i al be ween packe s. I should also be no ed ha hese esul s only show
compa isons o he wo highes p io i y slices, because he o he slices ha e e y high
ji e alues—mo e han 100 ms when highe p io i y slices demand bandwid h—wi h
high a ia ions, due o all he sa u a ion momen s hey su e . The e o e, we conside ed i
ele an o only show he ji e ob ained o he cases wi h mo e c i ical a ic.
Table 7p esen s he mean ji e alues ob ained nume ically o he Con ol slice,
highligh ing he in e als whe e he slice eques s a g ea e bandwid h, and Figu e 12 plo s
he mean, median, and de ia ion alues o he ji e ob ained in each 20 s in e al o he
Con ol slice in each o he h ee scena ios. I can be seen ha all o he alues shown a e
low, ne e eaching 3.5 ms o ji e . Rega ding he i s scena io, i can be obse ed ha
du ing he in e al whe e he Con ol use demands mo e bandwid h, he ji e wo sens,
inc easing i s mean and de ia ion alues and eaching alues o 2.435 ms, bu i emains
s able h oughou he es o he es , wi h mos o i s mean alues be ween 0.8 and 0.9 ms.
Howe e , Figu e 12 shows ha he e a e many a ypical poin s ou side he ange o he
de ia ion du ing he whole es . Conce ning he second scena io, i.e., he i s app oach,
he highes median, mean, and de ia ion alues a e ob ained, wi h a ypical poin s highe
han 2.5 ms. This leads o uns able beha io , especially when i is compa ed wi h o he
scena ios, whe e mos o he mean alues a e be ween 1.4 and 1.7 ms. Howe e , his
ange o alues can s ill be conside ed s able. Addi ionally, du ing he pe iod when mo e
bandwid h is needed, i can be seen ha he p io i iza ion ca ied ou educed he mean
alues o ji e in each in e al by app oxima ely 0.83 ms. Finally, he las scena io, i.e.,
he second app oach, shows ha he alues a e e y s able h oughou he es , wi h mos
o he alues main ained be ween 1.1 and 1.2 ms. As wi h he i s app oach, when he
g ea es bandwid h is eques ed, he ji e is educed in each in e al, ob aining he bes
esul s among all scena ios.
Elec onics 2022,11, 1097 21 o 27
Table 7. Mean ji e alues measu ed in each scena io in he Con ol slice.
Times
(s)
Wi hou
P io i iza ion
(ms)
Fi s
App oach
(ms)
Second
App oach
(ms)
0–20 0.86865 1.69075 0.8868
20–40 1.3488 1.3746 0.8973
40–60 1.75745 1.21645 0.8373
60–80 2.1286 0.9007 0.7713
80–100 2.43515 0.8378 0.79385
100–120 0.85055 1.5392 0.9122
120–140 0.83605 1.6702 1.03025
140–160 0.86075 1.4451 1.1007
160–180 0.9496 1.4847 1.1278
180–200 0.8391 1.70245 1.20035
200–220 0.82355 1.30005 1.1501
220–240 0.85865 1.48455 1.1113
240–260 0.83575 1.50085 1.108
260–280 0.83655 1.7163 1.13495
280–300 0.89005 1.64653 1.25858
Figu e 12. Compa ison o he ji e ob ained be ween he app oaches conside ed (Con ol a ic slice).
As in he p e ious case, Table 8p esen s he mean ji e alues ob ained nume ically in
he S eaming slice, highligh ing he in e als whe e he slice eques s a g ea e bandwid h,
and Figu e 13 shows he mean, median, and de ia ion alues o he ji e o he S eaming
slice. In he i s scena io, i can be seen ha he alues emain s able, wi h alues be ween
0.8 and 0.9, and inc ease sligh ly in he in e als co esponding o he momen o sa u a ion
o he slice. On he o he hand, in bo h he i s and he second app oaches, hese alues a e
s able h oughou he whole es —in he same way as in he Con ol slice and aking in o
accoun ha he mean, median, and a iance alues a e wide in he i s app oach—wi h
alues o he second scena io be ween 1.3 and 1.9 ms and alues o he hi d scena io
be ween 1 and 1.2 ms. The only momen when i can be seen ha bo h app oaches lose
s abili y and ha e much highe mean ji e alues is in he in e al whe e he highes
p io i y slice, i.e., he Con ol slice, eques s p ac ically all o he a ailable bandwid h and
he e o e a ec s he S eaming slice (80–100 s in e al). Finally, i is wo h no ing ha
du ing he pe iod in which mo e bandwid h is needed, ha he mean ji e alues in each
in e al a e also educed.
Elec onics 2022,11, 1097 22 o 27
Table 8. Mean ji e alues measu ed in each scena io in he S eaming slice.
Times
(s)
Wi hou
P io i iza ion
(ms)
Fi s
App oach
(ms)
Second
App oach
(ms)
0–20 0.8396 1.681 0.8614
20–40 0.8445 1.65525 0.9488
40–60 0.80605 1.92275 1.0087
60–80 0.82645 1.69285 1.3972
80–100 0.94615 2.8613 5.6984
100–120 0.90295 1.3117 0.89665
120–140 1.2251 1.17465 0.88205
140–160 1.7204 1.0637 0.84345
160–180 1.8813 1.39505 0.81155
180–200 0.8682 1.58875 1.13335
200–220 0.82765 1.37105 1.03115
220–240 0.83425 1.5474 1.05305
240–260 0.8319 1.6727 1.12325
260–280 0.80655 1.33285 1.0661
280–300 0.86074 1.37711 1.20058
Figu e 13.
Compa ison o he ji e ob ained be ween he app oaches conside ed (S eaming a ic slice).
6. Discussion
Taking in o accoun he esul s ob ained in Sec ion 5, a compa ison was made be ween
he h ee scena ios conside ed.
Be o e going in o he de ails, we show how he use o he ne wo k slicing concep
clea ly ob ains be e esul s in e ms o gua an eeing he ansmission o c i ical a ic such
as Con ol a ic, which co esponds o he highes p io i y slice in his pape . I is clea ly
seen ha , ega dless o he ne wo k slicing app oaches p oposed in his pape —scena ios
in which a single UE is associa ed wi h a single slice ( i s app oach) o scena ios in which
each UE is connec ed o all a ailable slices (second app oach)— he esul is sa is ac o y o
he highe p io i y slices. This applies o he h ee me ics ob ained, which a e bandwid h,
packe loss, and ji e .
I we ocus on he ob ained me ics, i is clea ha i we seek o gua an ee ha all
ypes o a ic, ega dless o hei cha ac e is ics, can ob ain an equi able amoun o a ic
ansmi ed, he scena io in which no ne wo k slicing app oach is con empla ed is he mos
sui able. Despi e gua an eeing bandwid h o all UEs in ol ed in he a ic ansmission,
he pe cen age o packe losses and ji e will ine i ably be high in hose UEs ha wan
o ansmi in o ma ion ha exceeds he sa u a ion le el o he ne wo k. Tha is why his
Elec onics 2022,11, 1097 23 o 27
app oach is no alid o indus ial o au omo i e use cases, which equi e he mos c i ical
da a o be ecei ed co ec ly and wi h educed ji e , ega dless o o he ypes o lowe
p io i y a ic ying o be sen [52,53].
The e o e, in hese ypes o applica ions o use cases, in which he sa is ac o y sending
o c i ical a ic p e ails o e non-p io i y a ic, he e is a bene i om he use o ei he
o he wo ne wo k slicing app oaches p oposed in his pape , as obse ed in he esul s.
Bo h ne wo k slicing app oaches a e able o p io i ize he highes p io i y a ic o e he
es o he a ic. Ne e heless, in all senses, i.e., in he maximum bandwid h ob ained, he
pe cen age o packe loss, and he amoun o ji e , he second app oach ob ained be e
esul s. I should also be emembe ed ha he i s app oach is no e ec i e in si ua ions
whe e he e is p io i y a ic bu wi h a small numbe o packe s, as he a ic may be
hidden unde he low p io i y bu dominan ype o a ic. Once again, o his ype o
applica ion, he second app oach is he mos e ec i e, because a UE is conside ed o be a
gene ic modem/ga eway, which ecei es di e en ypes o a ic om ex e nal sou ces
and is able o ea hem sepa a ely by co ec ly assigning he co esponding slice o he
a ic ca ego y.
Taking his in o accoun , ega ding bandwid h, e en hough he a ic ansmi ed
based on he highes p io i y slice in he wo ne wo k slicing app oaches p oposed in his
pape manages o each almos 105 Mbps, which is he maximum bandwid h p o ided
by ou expe imen al 5G SA ne wo k— he scena io wi hou ne wo k slicing does no e en
each 30 Mbps—in he case o he i s app oach, he es o he slices do no each he
maximum bandwid h possible. Fo example, he second slice, co esponding o S eaming
a ic, only eaches 60 Mbps in he i s app oach, compa ed wi h he 80 Mbps eached
by his slice in he second app oach. This di e ence be ween he bandwid hs ob ained is
e en g ea e o he o he lowe p io i y slices, which is a a o able esul in he case o he
second app oach.
Rega ding he packe losses, he mos ema kable obse a ion ob ained om he
compa ison o hese h ee scena ios is ha , wi h he second app oach, no packe is los in
he highes p io i y slice h oughou he es . Howe e , o he scena io whe e ne wo k
slicing is no conside ed and o he i s app oach, his is no he case.
Finally, he ji e esul s ob ained also show he supe io i y o he second app oach
wi h espec o he o he scena ios conside ed. The ji e alues ob ained in his case a e
e y s able h oughou he es , wi hou being a ec ed by sa u a ion and wi hou exceeding
1.26 ms ji e a any poin o he highes p io i y slice.
7. Conclusions
This pape e alua es he po en ial o ne wo k slicing applied o UL a RAN le el in a
dynamic en i onmen wi h mul iple da a lows and di e en equi emen s om di e en
ex e nal sou ces. Fo his pu pose, wo app oaches we e p oposed o gene a e a dynamic,
adap i e, and anspa en slicing o he end use s based on a ic classi ica ion. Fo bo h
cases, a a ic classi ica ion mechanism was ini ially used o classi y each ype. In a second
s ep, in he i s app oach his in o ma ion was used o ob ain he dominan a ic and
o ansmi all lows h ough he mos sui able slice o his dominan a ic. In con as ,
in he second app oach his in o ma ion was used o ob ain he classi ica ion o each o
he da a lows and o edi ec hem independen ly h ough he slice associa ed wi h ha
ca ego y. Howe e , i espec i e o whe he he i s o he second app oach was used,
his classi ica ion and he use o slices we e anspa en o he ex e nal sou ces sending
a ic o he UE, which ac s as a ga eway, ega dless o he ype o a ic. The algo i hm
gene a ed was esponsible o classi ica ion, slice selec ion, and ansmission o a ic o
he 5G ne wo k.
To e alua e he pe o mance o such app oaches, his pape deployed an expe imen al
es bed o a 5G SA ne wo k.
The esul s ob ained om hese app oaches, along wi h a e e ence scena io in which
no ne wo k slicing was conside ed, showed ha he use o ne wo k slicing echniques
Elec onics 2022,11, 1097 24 o 27
allows o highe e iciency and eliabili y o he mos c i ical da a, compa ed wi h hose
ha do no equi e gua an ees in e ms o equi emen s, such as packe loss o ji e .
In addi ion, i can be seen ha he second p oposed app oach showed an ad an age
o e he i s by ob aining g ea e con ol and accu acy in scena ios wi h he e ogeneous
a ic, by ea ing each ype o a ic ha passed h ough he UE independen ly, allowing
o he highes p io i y da a lows o be ecei ed co ec ly e en in si ua ions whe e he e oge-
neous a ic domina es. Howe e , he i s app oach can be mo e use ul when he ne wo ks
do no ha e many esou ces, such as in ne wo ks wi h ew a ailable IP add esses o in
scena ios whe e he ype o a ic is no known bu he a ic is known o be homogeneous.
Fu u e lines o esea ch should ocus on he applica ion o hese app oaches o end-
o-end ne wo k slicing o ensu e highe eliabili y h oughou he 5G ne wo k. Likewise,
be e isola ion o adio esou ces will also be sough , in o de o a oid sha ing esou ces
be ween slices. The e o e, each ca ego y o a ic will ha e i s own independen ne wo k.
Finally, we will seek o imp o e he dynamic selec ion o slices by means o machine
lea ning echniques.
Au ho Con ibu ions:
Concep ualiza ion and me hodology, Á.G., Z.F. and R.V.; so wa e and
in es iga ion, Á.G. and Z.F.; w i ing—o iginal d a p epa a ion, Á.G. and Z.F.; w i ing— e iew and
edi ing, Á.G., Z.F., R.V., Á.M., M.Z., P.A. and J.M. All au ho s ha e ead and ag eed o he published
e sion o he manusc ip .
Funding:
This esea ch was suppo ed by he Spanish Cen e o he De elopmen o Indus ial
Technology (CDTI) and he Minis y o Economy, Indus y and Compe i i eness unde g an /p ojec
CER-20191015/Open, Vi ualized Technology Demons a o s o Sma Ne wo ks (Open-VERSO).
Da a A ailabili y S a emen : Expe imen al da a a e a ailable upon eques .
Acknowledgmen s:
This esea ch was suppo ed by he Spanish Cen e o he De elopmen o
Indus ial Technology (CDTI) and he Minis y o Economy, Indus y and Compe i i eness unde
g an /p ojec CER-20191015/Open, Vi ualized Technology Demons a o s o Sma Ne wo ks
(Open-VERSO) and he Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og amme unde
g an ag eemen No. 957360 (5GMETA p ojec ).
Con lic s o In e es : The au ho s decla e no con lic o in e es .
Abb e ia ions
The ollowing abb e ia ions a e used in his manusc ip :
3GPP 3 d Gene a ion Pa ne ship P ojec
5GC 5G Co e
5GPPP 5G In as uc u e Public P i a e Pa ne ship
5QI 5G QoS iden i ie
DL Downlink
DPI Deep packe inspec ion
eMBB Enhanced mobile b oadband
gNB gNodeB
HMTC High-pe o mance machine- ype communica ions
HTTP Hype ex T ans e P o ocol
ICMP In e ne Con ol Message P o ocol
IoT In e ne o Things
IIoT Indus ial In e ne o Things
KPI Key pe o mance indica o s
MAC Media access con ol
MEC Mul i-access edge compu ing
MIoT Massi e IoT
mMTC Massi e machine- ype communica ions
MQTT Message Queue Teleme y T anspo
NGAP NG Applica ion P o ocol
Elec onics 2022,11, 1097 25 o 27
NGMN Nex Gene a ion Mobile Ne wo k
NFV Ne wo k unc ion i ualiza ion
OAI OpenAi In e ace
OSI Open sys ems in e connec ion
PDU P o ocol da a uni
PRB Physical esou ce block
QoS Quali y o se ice
RAN Radio access ne wo k
RTP Real Time P o ocol
SA S andalone
SD Slice di e en ia o
SDN So wa e-de ined ne wo k
SPI S ochas ic packe inspec ion
SSL Secu e socke s laye
SST Slice/se ice ype
S-NSSAI Single Ne wo k Slice Selec ion Assis ance In o ma ion
TCU Telema ic con ol uni
UDP Use Da ag am P o ocol
UE Use equipmen
UL Uplink
URLLC Ul a- eliable low-la ency communica ions
V2X Vehicle o e e y hing
VNF Vi ualized ne wo k unc ions
Re e ences
1.
Shu, Z.; Taleb, T. A no el QoS amewo k o ne wo k slicing in 5G and beyond ne wo ks based on SDN and NFV. IEEE Ne w.
2020,34, 256–263. [C ossRe ]
2.
Sohaib, R.M.; Oni e i, O.; Sambo, Y.; Im an, M.A. Ne wo k Slicing o Beyond 5G Sys ems: An O e iew o he Sma Po Use
Case. Elec onics 2021,10, 1090. [C ossRe ]
3.
Cisco, U. Cisco Annual In e ne Repo (2018–2023) Whi e Pape . 2020. A ailable online: h ps://www.cisco.com/c/en/us/
solu ions/colla e al/execu i e-pe spec i es/annual-in e ne - epo /whi e-pape -c11-741490.h ml (accessed on 15 Janua y 2022).
4. 5GPPP: The 5G In as uc u e Public P i a e Pa ne ship. A ailable online: h ps://5g-ppp.eu/ (accessed on 7 Janua y 2022).
5.
5G-ACIA: 5G Alliance o Connec ed Indus ies and Au oma ion. A ailable online: h ps://5g-acia.o g/ (accessed on 7 Janua y 2022).
6. 5G-AA: 5G Au omo i e Associa ion. A ailable online: h ps://5gaa.o g/ (accessed on 7 Janua y 2022).
7.
Siddiqi, M.A.; Yu, H.; Joung, J. 5G ul a- eliable low-la ency communica ion implemen a ion challenges and ope a ional issues
wi h IoT de ices. Elec onics 2019,8, 981. [C ossRe ]
8.
Rinaldi, F.; Raschella, A.; Pizzi, S. 5G NR sys em design: A concise su ey o key ea u es and capabili ies. Wi el. Ne w.
2021
,
27, 5173–5188. [C ossRe ]
9.
Teng, Y.; Yan, M.; Liu, D.; Han, Z.; Song, M. Dis ibu ed lea ning solu ion o uplink a ic con ol in ene gy ha es ing massi e
machine- ype communica ions. IEEE Wi el. Commun. Le . 2019,9, 485–489. [C ossRe ]
10.
3GPP. Sys em A chi ec u e o he 5G Sys em (5GS). TS 23.501 V17.2.0. Sep embe 2021. A ailable online: h ps://www.3gpp.
o g/ p/Specs/a chi e/23_se ies/23.501/23501-h20.zip (accessed on 22 Decembe 2021).
11.
NGMN Alliance. 5G Whi e Pape . In Nex Gene a ion Mobile Ne wo ks, Whi e Pape ; NGMN Alliance: F ank u am Main, Ge many,
2015; Volume 1. A ailable online: h ps://ngmn.o g/wp-con en /uploads/NGMN_5G_Whi e_Pape _V1_0.pd (accessed on 20
Janua y 2022).
12.
A olabi, I.; Taleb, T.; Samdanis, K.; Ksen ini, A.; Flinck, H. Ne wo k slicing and so wa iza ion: A su ey on p inciples, enabling
echnologies, and solu ions. IEEE Commun. Su . Tu o . 2018,20, 2429–2453. [C ossRe ]
13.
Ba akabi ze, A.A.; Ahmad, A.; Mijumbi, R.; Hines, A. 5G ne wo k slicing using SDN and NFV: A su ey o axonomy, a chi ec u es
and u u e challenges. Compu . Ne w. 2020,167, 106984. [C ossRe ]
14.
A olabi, I.; Taleb, T.; F angoudis, P.A.; Bagaa, M.; Ksen ini, A. Ne wo k slicing-based cus omiza ion o 5G mobile se ices. IEEE
Ne w. 2019,33, 134–141. [C ossRe ]
15.
G anelli, F. Chap e 3-Ne wo k slicing. In Compu ing in Communica ion Ne wo ks; Fi zek, F.H., G anelli, F., Seeling, P., Eds.;
Academic P ess: Camb idge, MA, USA, 2020; pp. 63–76. [C ossRe ]
16.
Elayoubi, S.E.; Jemaa, S.B.; Al man, Z.; Galindo-Se ano, A. 5G RAN slicing o e icals: Enable s and challenges. IEEE Commun.
Mag. 2019,57, 28–34. [C ossRe ]
17.
5G-NORMA—5G No el Radio Mul ise ice Adap i e Ne wo k A chi ec u e. A ailable online: h ps://www.i .uc3m.es/wnl/
5gno ma/ (accessed on 7 Janua y 2022).
18.
SLICENET: End- o-End Cogni i e Ne wo k Slicing and Slice Managemen F amewo k in Vi ualised Mul i-Domain, Mul i-Tenan
5G Ne wo ks. A ailable online: h ps://slicene .eu/ (accessed on 7 Janua y 2022).
