Co esponding au ho : Gokul Chand a Pu nachand a Reddy
Copy igh © 2025 Au ho (s) e ain he copy igh o his a icle. This a icle is published unde he e ms o he C ea i e Commons A ibu ion Liscense 4.0.
Building esilien hyb id clouds: A laye ed a chi ec u e o aul ole ance and
pe o mance op imiza ion
Gokul Chand a Pu nachand a Reddy *
Amazon Web Se ices (AWS), USA.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2146-2152
Publica ion his o y: Recei ed on 30 Ma ch 2025; e ised on 09 May 2025; accep ed on 11 May 2025
A icle DOI: h ps://doi.o g/10.30574/wja .2025.26.2.1750
Abs ac
This a icle explo es he e olu ion o ne wo k in as uc u e om adi ional ha dwa e-based appliances o so wa e-
de ined solu ions, ocusing on he in eg a ion o p og ammable ha dwa e accele a o s wi hin Kube ne es-o ches a ed
en i onmen s. The a icle examines how Vi ual Ne wo k Func ions (VNFs) ha e ans o med se ice deli e y
capabili ies while add essing pe o mance challenges h ough ha dwa e accele a ion echnologies including FPGAs,
GPUs, and DPUs. The a icle in es iga es he ole o Kube ne es as an o ches a ion ounda ion, analyzing i s De ice
Plugin amewo k, Cus om Resou ce De ini ions, and Ne wo k Se ice Mesh in eg a ion. Th ough de ailed pe o mance
analysis, he a icle demons a es how p ope o ches a ion s a egies and esou ce op imiza ion echniques can
enhance sys em e iciency, educe ope a ional cos s, and imp o e se ice quali y in hyb id cloud en i onmen s. The
a icle also explo es u u e de elopmen s in cloud-na i e compu ing, including disagg ega ed da acen e a chi ec u es
and edge compu ing in eg a ion, p o iding insigh s in o he nex gene a ion o ne wo k in as uc u e solu ions.
Keywo ds: Ha dwa e Accele a ion; Vi ual Ne wo k Func ions; Kube ne es O ches a ion; Edge Compu ing; Cloud-
Na i e In as uc u e
1. In oduc ion
The ans o ma ion o ne wo k in as uc u e om ha dwa e-based appliances o so wa e-de ined solu ions
ep esen s a undamen al shi in how ne wo k se ices a e deli e ed and managed. Resea ch by Mijumbi e al.
demons a es ha Vi ual Ne wo k Func ions (VNFs) ha e eme ged as a c i ical echnology o educing bo h capi al
and ope a ional expendi u es in ne wo k deploymen s, wi h o ganiza ions achie ing up o 30% educ ion in o e all
in as uc u e cos s compa ed o adi ional ha dwa e-based solu ions [1].
The adop ion o VNFs has been d i en by hei abili y o decouple ne wo k unc ions om p op ie a y ha dwa e
pla o ms. Acco ding o comp ehensi e analysis in he IEEE Communica ions Su eys & Tu o ials, his decoupling
enables ne wo k ope a o s o deploy se ices up o 8 imes as e han adi ional ha dwa e-based app oaches, while
main aining he lexibili y o scale esou ces based on ac ual demand [1]. Howe e , he ansi ion has in oduced new
challenges, pa icula ly in main aining pe o mance pa i y wi h dedica ed ha dwa e solu ions.
Pe o mance op imiza ion in NFV en i onmen s has become a c i ical ocus a ea, especially as o ganiza ions mo e
owa d hyb id cloud a chi ec u es. Resea ch by Ka sikas e al. demons a es ha adi ional so wa e-based NFV
implemen a ions can expe ience signi ican pe o mance deg ada ion, wi h packe p ocessing o e head inc easing by
up o 4.5 imes compa ed o ba e-me al solu ions [2]. Thei wo k in de eloping he Me on a chi ec u e showed ha by
in eg a ing ha dwa e accele a ion and op imizing packe p ocessing pa hs, i 's possible o achie e pe o mance le els
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2146-2152
2147
ha ma ch dedica ed ha dwa e solu ions, p ocessing up o 100 Gbps o a ic while main aining consis en la ency
p o iles [2].
The in eg a ion o p og ammable ha dwa e accele a o s wi hin Kube ne es-o ches a ed en i onmen s has p o en
pa icula ly e ec i e in add essing hese pe o mance challenges. The Me on amewo k, as de ailed in he USENIX
Symposium p esen a ion, demons a ed ha ha dwa e-accele a ed VNFs can achie e p ocessing speeds o up o 82
million packe s pe second while main aining sub-100 mic osecond la ency, ep esen ing pe o mance le els
p e iously only achie able wi h dedica ed ha dwa e solu ions [2]. These imp o emen s a e ealized h ough
sophis ica ed packe p ocessing op imiza ions and he e icien u iliza ion o ha dwa e accele a ion capabili ies.
In hyb id cloud deploymen s, he pe o mance bene i s o ha dwa e accele a ion become e en mo e p onounced. The
esea ch p esen ed a NSDI '18 shows ha p ope ly o ches a ed ha dwa e-accele a ed VNFs can main ain consis en
pe o mance ac oss di e se in as uc u e en i onmen s, wi h packe p ocessing e iciency eaching up o 97% o he
heo e ical maximum o he unde lying ha dwa e [2]. This le el o pe o mance op imiza ion is c ucial o
o ganiza ions deploying complex ne wo k se ices ac oss hyb id cloud en i onmen s.
2. The E olu ion o Ne wo k Se ices
The e olu ion o ne wo k se ices ep esen s a undamen al shi in how elecommunica ions in as uc u e is deployed
and managed. Resea ch published in Science Di ec demons a es ha adi ional ha dwa e-based deploymen s
his o ically consumed signi ican po ions o IT budge s, wi h ne wo k unc ion-speci ic ha dwa e accoun ing o
app oxima ely 60% o o al in as uc u e cos s in en e p ise en i onmen s [3]. The mig a ion o so wa e-de ined
solu ions has enabled o ganiza ions o educe hese ha dwa e- ela ed expenses while main aining se ice quali y and
eliabili y.
The ans o ma ion o Vi ual Ne wo k Func ions (VNFs) has e olu ionized se ice deli e y capabili ies in bo h
en e p ise and se ice p o ide ne wo ks. Pe o mance analysis o VNF deploymen s has shown ha i ualized
ne wo k unc ions can achie e h oughpu a es o up o 8.5 Gbps pe ins ance when p ope ly op imized, hough his
a ies signi ican ly based on he speci ic unc ion being i ualized and he unde lying ha dwa e pla o m [3]. This
ep esen s a no able achie emen in he e olu ion o ne wo k se ices, demons a ing ha so wa e-based solu ions
can deli e subs an ial pe o mance while o e ing g ea e lexibili y han adi ional ha dwa e appliances.
The implemen a ion o se ice unc ion chaining in i ualized en i onmen s has in oduced new conside a ions o
pe o mance op imiza ion. Acco ding o esea ch ocused on ne wo k se ice unc ion chaining, deploymen s u ilizing
op imal opology con igu a ions can achie e end- o-end la ency as low as 125 mic oseconds o basic se ice chains,
hough his inc eases o app oxima ely 275-350 mic oseconds as addi ional ne wo k unc ions a e added o he chain
[4]. These measu emen s p o ide c ucial insigh s in o he eal-wo ld pe o mance cha ac e is ics o so wa e-de ined
ne wo king solu ions.
Pe o mance analysis o compu e-in ensi e asks in VNF en i onmen s e eals speci ic pa e ns in esou ce u iliza ion
and p ocessing capabili ies. S udies show ha deep packe inspec ion unc ions in i ualized en i onmen s can inspec
and p ocess a ic a a es o up o 3.2 Gbps pe CPU co e, hough his a e dec eases by app oxima ely 45% when
dealing wi h enc yp ed a ic lows [4]. This esea ch highligh s bo h he capabili ies and limi a ions o cu en
so wa e-de ined ne wo king app oaches, pa icula ly in scena ios equi ing in ensi e packe p ocessing and analysis.
Table 1 Compa a i e E iciency Me ics in Pe cen ages [3, 4]
E iciency Ca ego y
T adi ional In as uc u e
VNF En i onmen
Resou ce U iliza ion
60%
85%
P ocessing Capaci y
70%
95%
Cos E iciency
40%
70%
Pe o mance Scaling
55%
85%
Ope a ional E iciency
50%
80%
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2146-2152
2148
3. Le e aging P og ammable Ha dwa e Accele a o s
The in eg a ion o p og ammable ha dwa e accele a o s has become c ucial o add essing pe o mance challenges in
i ualized ne wo k en i onmen s. Resea ch on P4 o FPGA implemen a ions demons a es ha ha dwa e accele a ion
can achie e p ocessing a es o up o 40 Gbps o complex packe p ocessing asks, wi h expe imen al esul s showing
la ency educ ions o up o 70% compa ed o so wa e-based solu ions [5]. These pe o mance imp o emen s a e
pa icula ly signi ican in scena ios equi ing eal- ime packe p ocessing and analysis.
Field-P og ammable Ga e A ays (FPGAs) ha e demons a ed excep ional capabili ies in ne wo k unc ion accele a ion,
pa icula ly when implemen ed wi h P4-based p og amming amewo ks. Pe o mance analysis shows ha FPGA-
based packe p ocesso s can achie e h oughpu a es o 10-40 Gbps while main aining consis en la ency p o iles
unde 100 nanoseconds pe packe [5]. The s udy e ealed ha hese implemen a ions can handle complex packe
p ocessing asks wi h esou ce u iliza ion emaining below 35% o a ailable FPGA logic elemen s, enabling mul iple
ne wo k unc ions o be implemen ed on a single de ice.
G aphics P ocessing Uni s (GPUs) ha e e olu ionized pa allel p ocessing capabili ies in ne wo k en i onmen s.
Acco ding o he Packe Shade esea ch, GPU-accele a ed so wa e ou e s can achie e packe p ocessing a es o up o
39 Gbps o IP 4 packe s and 32 Gbps o IP 6 packe s using a single GPU [6]. The s udy demons a ed ha GPU-based
implemen a ions can p ocess IP 4 packe s a a es exceeding 4 million packe s pe second, ep esen ing a signi ican
ad ancemen in so wa e-based ou ing pe o mance.
Da a P ocessing Uni s (DPUs) and hei in eg a ion wi h GPU accele a ion ha e shown ema kable e iciency gains in
ne wo k p ocessing asks. Resea ch indica es ha GPU-accele a ed implemen a ions can educe CPU co e u iliza ion
om 100% o app oxima ely 30% while main aining ull line- a e p ocessing capabili ies [6]. These imp o emen s a e
pa icula ly no able in IPsec p ocessing, whe e GPU accele a ion enables h oughpu a es o up o 21.2 Gbps wi h AES-
128 enc yp ion, ep esen ing a 3.5x pe o mance imp o emen o e CPU-only implemen a ions.
The o e all impac o ha dwa e accele a ion ex ends o sys em-wide pe o mance imp o emen s. Implemen a ion
s udies show ha GPU-accele a ed packe p ocessing can achie e up o 4x pe o mance imp o emen in ou ing asks
while educing powe consump ion p opo ionally o he numbe o CPU co es ha can be idled [6]. This e iciency gain
is pa icula ly signi ican in la ge-scale deploymen s whe e powe consump ion and p ocessing capabili y mus be
ca e ully balanced.
4. Kube ne es as an O ches a ion Founda ion
The adop ion o Kube ne es as an o ches a ion pla o m o Vi ual Ne wo k Func ions (VNFs) ep esen s a signi ican
ad ancemen in con aine -based ne wo k managemen . Pe o mance analysis compa ing i ual machines and Linux
con aine s e eals ha con aine -based deploymen s can achie e s a up imes app oxima ely 5 imes as e han
equi alen i ual machine implemen a ions, wi h con aine c ea ion and des uc ion ope a ions comple ing in unde
2 seconds compa ed o 10-15 seconds o i ual machines [7]. These pe o mance cha ac e is ics make Kube ne es
pa icula ly sui able o dynamic ne wo k unc ion o ches a ion.
The De ice Plugin amewo k in Kube ne es demons a es signi ican ad an ages in esou ce managemen e iciency.
Resea ch indica es ha con aine ized applica ions show subs an ially lowe memo y o e head, using only ens o
megaby es compa ed o hund eds o megaby es equi ed by i ual machines o equi alen wo kloads [7]. This
e iciency ex ends o CPU u iliza ion, whe e con aine -based ne wo k unc ions exhibi app oxima ely 2% o e head
compa ed o ba e-me al pe o mance, making hem highly sui able o ha dwa e-accele a ed deploymen s.
Ne wo k pe o mance op imiza ion s udies ocusing on high-pe o mance ne wo ks ha e e ealed subs an ial
imp o emen s in esou ce u iliza ion h ough Cus om Resou ce De ini ions (CRDs). Analysis shows ha p ope ly
implemen ed ne wo k op imiza ion models can achie e up o 95% ne wo k u iliza ion e iciency while main aining
la ency wi hin accep able bounds [8]. The esea ch demons a es ha CRD-based managemen can handle ne wo k
con igu a ion changes wi h esponse imes a e aging 40-50 milliseconds, enabling apid adap a ion o changing
ne wo k condi ions.
Ne wo k Se ice Mesh (NSM) in eg a ion has p o en pa icula ly e ec i e in enhancing ne wo k pe o mance in
con aine ized en i onmen s. S udies o op imiza ion models o high-pe o mance ne wo ks indica e ha se ice mesh
implemen a ions can educe c oss-clus e communica ion la ency by up o 40% compa ed o adi ional ou ing
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2146-2152
2149
app oaches [8]. The esea ch shows ha op imized se ice mesh con igu a ions can main ain consis en pe o mance
while handling up o 1,000 simul aneous se ice connec ions, wi h packe p ocessing o e head emaining below 5% o
o al sys em esou ces.
Pe o mance analysis o in eg a ed Kube ne es en i onmen s e eals signi ican ad an ages in esou ce u iliza ion and
ope a ional e iciency. Con aine -based ne wo king solu ions show memo y oo p in s app oxima ely 60% smalle
han adi ional i ual machine deploymen s, while main aining equi alen unc ionali y [7]. These e iciency gains a e
pa icula ly impo an in en i onmen s whe e ha dwa e accele a ion is employed, as hey enable mo e esou ces o be
dedica ed o ac ual packe p ocessing a he han sys em o e head.
Table 2 Ha dwa e Accele a o E iciency Imp o emen s [7, 8]
Pe o mance Me ic
Base Sys em
Imp o emen
FPGA La ency E iciency
30%
70%
FPGA Resou ce E iciency
65%
35%
GPU IP 4 P ocessing
25%
75%
GPU IP 6 P ocessing
20%
80%
GPU CPU U iliza ion
30%
70%
IPsec Pe o mance
29%
71%
Powe E iciency
25%
75%
5. O ches a ion S a egies and Bes P ac ices
E ec i e o ches a ion o ha dwa e-accele a ed Vi ual Ne wo k Func ions (VNFs) demands sophis ica ed esou ce
managemen and moni o ing s a egies. Resea ch on con aine o ches a ion in mul i-cloud en i onmen s
demons a es ha op imized esou ce alloca ion can imp o e esou ce u iliza ion by up o 30% while educing
deploymen cos s by app oxima ely 25% compa ed o non-op imized deploymen s [9]. These imp o emen s a e
achie ed h ough s a egic wo kload placemen and esou ce dis ibu ion, wi h s udies showing ha in elligen
o ches a ion can educe o e all esou ce consump ion by up o 20% in mul i- enan en i onmen s.
Resou ce managemen s a egies ha conside mic ose ices-based a chi ec u es ha e shown signi ican ope a ional
bene i s. Analysis e eals ha con aine o ches a ion op imiza ion can educe se ice esponse imes by up o 40%
while imp o ing o e all sys em h oughpu by 35% compa ed o baseline implemen a ions [9]. In dis ibu ed
en i onmen s, hese op imiza ion app oaches ha e demons a ed he abili y o educe in e -se ice communica ion
o e head by app oxima ely 28%, while main aining consis en pe o mance ac oss mul iple cloud p o ide s.
Vi ual ne wo k esou ce op imiza ion esea ch has e ealed subs an ial imp o emen s h ough in elligen esou ce
alloca ion models. S udies indica e ha op imized VNF placemen can achie e up o 92% esou ce u iliza ion e iciency
while educing ope a ional cos s by app oxima ely 33% [10]. The esea ch demons a es ha p ope esou ce
alloca ion s a egies can main ain ne wo k se ice quali y while educing he o al numbe o equi ed ne wo k nodes
by up o 25%, pa icula ly in scena ios in ol ing mul iple se ice chains.
Moni o ing and pe o mance op imiza ion in i ualized en i onmen s plays a c ucial ole in main aining se ice
quali y. Analysis shows ha implemen ing esou ce op imiza ion models can imp o e ne wo k h oughpu by up o
45% while educing end- o-end se ice la ency by app oxima ely 30% [10]. The esea ch indica es ha op imized
esou ce alloca ion can suppo up o 85% mo e concu en ne wo k se ices compa ed o adi ional deploymen
app oaches, while main aining consis en quali y o se ice me ics.
Implemen a ion o hese o ches a ion s a egies has shown measu able imp o emen s in o e all sys em e iciency.
S udies demons a e ha p ope ly op imized en i onmen s can achie e esou ce u iliza ion imp o emen s o up o
40% while educing ope a ional cos s by app oxima ely 35% ac oss dis ibu ed deploymen s [9]. This le el o
op imiza ion is achie ed h ough ca e ul conside a ion o se ice equi emen s, esou ce a ailabili y, and in e -se ice
dependencies, enabling e icien esou ce usage while main aining pe o mance objec i es.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2146-2152
2150
Table 3 Resou ce Op imiza ion and Cos E iciency Me ics [9, 10]
Op imiza ion Ca ego y
Baseline
Imp o emen
Resou ce U iliza ion
70%
30%
Deploymen Cos s
75%
25%
Resou ce Consump ion
80%
20%
Se ice Response Time
60%
40%
Sys em Th oughpu
65%
35%
Communica ion O e head
72%
28%
6. Looking Ahead: Fu u e De elopmen s
The e olu ion o ha dwa e accele a ion in cloud-na i e en i onmen s con inues o ad ance apidly, wi h eme ging
echnologies eshaping da acen e a chi ec u es. Resea ch on disagg ega ed da acen e s shows ha nex -gene a ion
ne wo k a chi ec u es can achie e bandwid h u iliza ion o up o 90% while main aining la ency unde 10
mic oseconds o mos ope a ions [11]. These ad ancemen s in ne wo k disagg ega ion ha e demons a ed he abili y
o educe o al cos o owne ship by up o 47% compa ed o adi ional da acen e designs, pa icula ly when
implemen ing p og ammable ne wo k in e aces and sma NICs.
Ne wo k pe o mance in disagg ega ed en i onmen s has shown ema kable imp o emen s h ough ad anced
esou ce managemen . S udies indica e ha disagg ega ed a chi ec u es can suppo up o 100 Gbps o bandwid h pe
node while main aining consis en la ency p o iles below 15 mic oseconds ac oss di e se wo kloads [11]. The esea ch
demons a es ha p ope ly implemen ed esou ce disagg ega ion can imp o e o e all sys em e iciency by up o 35%
while educing ha dwa e cos s by app oxima ely 40% compa ed o adi ional monoli hic a chi ec u es.
Edge compu ing in eg a ion has eme ged as a c i ical ocus a ea o u u e de elopmen . Resea ch on IoT edge
compu ing shows ha ligh weigh i ualiza ion echnologies can educe esou ce usage by up o 30% compa ed o
adi ional i ualiza ion app oaches [12]. Pe o mance analysis e eals ha con aine ized edge applica ions can
achie e s a up imes unde 50 milliseconds, while main aining CPU u iliza ion below 20% du ing no mal ope a ion,
ep esen ing signi ican imp o emen s o e con en ional i ual machine deploymen s.
The implemen a ion o ligh weigh i ualiza ion a he edge has demons a ed subs an ial bene i s o esou ce-
cons ained en i onmen s. S udies show ha op imized con aine deploymen s can educe memo y oo p in by up o
28% while imp o ing applica ion densi y by app oxima ely 2.5 imes compa ed o adi ional i ualiza ion solu ions
[12]. These imp o emen s enable edge nodes o suppo mo e concu en se ices while main aining consis en
pe o mance le els, pa icula ly impo an o IoT and edge compu ing scena ios.
Table 4 Edge Compu ing and Vi ualiza ion E iciency [11, 12]
Pe o mance Me ic
T adi ional VM
Imp o emen
Resou ce Usage E iciency
70%
30%
CPU U iliza ion
80%
60%
Memo y Foo p in E iciency
72%
28%
Applica ion Densi y
40%
60%
Resou ce U iliza ion
65%
35%
Se ice Concu ency
40%
60%
These de elopmen s collec i ely highligh he e ol ing landscape o cloud-na i e compu ing and ha dwa e accele a ion.
The esea ch indica es ha combining disagg ega ed a chi ec u es wi h ligh weigh i ualiza ion can educe o e all
sys em la ency by up o 45% while imp o ing esou ce u iliza ion by app oxima ely 35% [11, 12]. These ad ancemen s
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2146-2152
2151
sugges a u u e whe e highly e icien , disagg ega ed compu ing en i onmen s become he no m, enabling mo e
lexible and cos -e ec i e deploymen o ne wo k se ices.
7. Conclusion
The ans o ma ion o ne wo k in as uc u e h ough he adop ion o so wa e-de ined solu ions and ha dwa e
accele a ion has undamen ally changed how o ganiza ions deploy and manage ne wo k se ices. The in eg a ion o
p og ammable ha dwa e accele a o s wi hin Kube ne es-o ches a ed en i onmen s has success ully add essed he
pe o mance challenges inhe en in i ualized ne wo k unc ions while main aining he lexibili y and cos bene i s o
so wa e-de ined ne wo king. The implemen a ion o sophis ica ed o ches a ion s a egies, combined wi h ad ances
in esou ce managemen and moni o ing, has enabled o ganiza ions o achie e op imal pe o mance le els while
educing ope a ional complexi y. As he echnology con inues o e ol e, he eme gence o disagg ega ed a chi ec u es
and edge compu ing solu ions poin s owa d a u u e whe e highly e icien , lexible, and cos -e ec i e ne wo k se ices
become inc easingly accessible. This a icle demons a es ha he combina ion o ha dwa e accele a ion, con aine
o ches a ion, and in elligen esou ce managemen p o ides a obus ounda ion o building esilien hyb id cloud
a chi ec u es ha can mee he demanding equi emen s o mode n ne wo k se ices.
Re e ences
[1] Rashid Mijumbi, "Ne wo k Func ion Vi ualiza ion: S a e-o - he-A and Resea ch Challenges," IEEE
Communica ions Su eys & Tu o ials, ol. 18, no. 1, pp. 236-262, Sep embe 2015.
h ps://www. esea chga e.ne /publica ion/281524200_Ne wo k_Func ion_Vi ualiza ion_S a e-o - he-
A _and_Resea ch_Challenges
[2] Geogias Ka sikas e al. "Me on: NFV Se ice Chains a he T ue Speed o he Unde lying Ha dwa e," 2018. URL:
h ps://www.usenix.o g/con e ence/nsdi18/p esen a ion/ka sikas
[3] Alan Zeichik e al., "Enabling inno a ion by opening up he ne wo k," Science Di ec , ol. 7, no. 2, pp. 156-169,
Ap il 2017. h ps://www.sciencedi ec .com/science/a icle/abs/pii/S1353485817300387
[4] Gab iel A aujo, "Ne wo k Se ice Func ion Chaining: A Pe o mance S udy Va ying Topologies," IEEE Jou nal on
Selec ed A eas in Communica ions, ol. 40, no. 3, pp. 982-995, Oc obe 2024.
h ps://www. esea chga e.ne /publica ion/384937206_Ne wo k_se ice_ unc ion_chaining_a_pe o mance_s
udy_ a ying_ opologies
[5] Zuang Cao e al., "P4 o FPGA - A Fas App oach o Gene a ing E icien Ne wo k P ocesso s," 2020 In e na ional
Con e ence on Field-P og ammable Technology (ICFPT), Janua y 2020. URL:
h ps://www. esea chga e.ne /publica ion/338926337_P4_ o_FPGA_-
_A_Fas _App oach_ o _Gene a ing_E icien _Ne wo k_P ocesso s
[6] Sangjin Han, "Packe Shade : a GPU-Accele a ed So wa e Rou e ," ACM SIGCOMM Compu e Communica ion
Re iew, ol. 40, no. 4, pp. 195-206, 2010. URL: h ps://keonjang.gi hub.io/pape s/sigcomm10ps.pd
[7] Wes Fel e e al., "An upda ed pe o mance compa ison o i ual machines and Linux con aine s," IEEE
In e na ional Symposium on Pe o mance Analysis o Sys ems and So wa e (ISPASS), pp. 171-172, 2015. URL:
h ps://ieeexplo e.ieee.o g/documen /7095802
[8] Oluwa obi Akinme ese e al., "A S udy o Ne wo k Op imiza ion Models o High-Pe o mance Ne wo ks,"
In e na ional Jou nal o Ne wo k Managemen , ol. 33, no. 4, June 2023. URL:
h ps://www. esea chga e.ne /publica ion/371563708_A_S udy_o _Ne wo k_Op imiza ion_Models_ o _High-
Pe o mance_Ne wo ks
[9] Ca los Gue e o e al., "Resou ce Op imiza ion o Con aine O ches a ion: A Case S udy in Mul i-Cloud
Mic ose ices-based Applica ions," In e na ional Jou nal o Ne wo k Managemen , July 2018. URL:
h ps://www. esea chga e.ne /publica ion/324160574_Resou ce_op imiza ion_o _con aine _o ches a ion_a_c
ase_s udy_in_mul i-cloud_mic ose ices-based_applica ions
[10] Dani Vladisla ic e al., "Vi ual Ne wo k Resou ce Op imiza ion Model o Ne wo k Func ion Vi ualiza ion,"
In e na ional Con e ence on Communica ions and Ne wo king, Augus 2021. URL:
h ps://www. esea chga e.ne /publica ion/353967405_Vi ual_Ne wo k_Resou ce_Op imiza ion_Model_ o _N
e wo k_Func ion_Vi ualiza ion
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 26(02), 2146-2152
2152
[11] B ice Ekane e al., "Ne wo k in Disagg ega ed Da acen e s," IEEE/ACM T ansac ions on Ne wo king, Ma ch 2021.
URL: h ps://www. esea chga e.ne /publica ion/350808435_Ne wo k_in_Disagg ega ed_Da acen e s
[12] Robe o Mo abi o, "Consolida e IoT Edge Compu ing wi h Ligh weigh Vi ualiza ion," IEEE Ne wo k, ol. 31, no.
6, Janua y 2018. URL:
h ps://www. esea chga e.ne /publica ion/320729008_Consolida e_IoT_Edge_Compu ing_wi h_Ligh weigh _V
i ualiza ion
