In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen (IJTSRD)
Volume 9 Issue 6, No -Dec 2025 A ailable Online: www.ij s d.com e-ISSN: 2456 – 6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 186
Op imizing Ene gy Consump ion h ough
Hyb id Edge-Cloud Compu a ion Models
Awodi e, Moyosoluwa
1
; Ab ahams, Oluwa emi
2
1
IT Manage ,
S ockholm Uni e si y, S ockholm, Sweden
2
Senio Pla o m Enginee , Lau ea Uni e si y o Applied Sciences, Van aa, Finland
ABSTRACT
The apidly inc easing numbe o In e ne o Things (IoT) de ices is
o ecas o exceed 75 billion by 2025, d i ing demand o ene gy-
e icien compu ing amewo ks o suppo da a-in ensi e
applica ions in eme ging echnologies such as sma ci ies,
heal hca e, and indus ial au oma ion. Edge-cloud compu ing
a chi ec u e le e ages bo h edge p ocessing capaci y and cen alized
cloud p ocessing capaci y o add ess he inhe en limi a ions o edge
p ocessing, namely ene gy cos s associa ed wi h limi ed esou ces a
he edge o ansmission cos s (including ene gy and delay)
associa ed wi h sending da a back and o h o he cloud. In his
pape , an Ene gy-Awa e Task O loading (EATO) algo i hm is
p oposed ha dynamically o loads asks o edge de ices, edge
se e s, and he cloud o op imized ene gy consump ion and quali y
o se ice (QoS). The EATO algo i hm u ilizes eal- ime ene gy
p o iling, ne wo k condi ions, and compu a ional equi emen s, and is
calcula ed as a ma hema ical op imiza ion p oblem. The EATO
algo i hm was e alua ed using a simula ion o 100 IoT de ices and
ound o educe ene gy consump ion by up o 25% compa ed o edge-
only and cloud-only app oaches, while p oducing a 21%
enhancemen in ask scheduling ime o e s a e-o - he-a me hods
[15]. The pape makes wo main con ibu ions: a gene ic, scalable
ask o loading amewo k and an examina ion o hyb id-based
a chi ec u e o sus ainable compu ing. The indings will encou age
esea che s o ocus on ene gy e iciency o IoT deploymen s, and
u u e wo k will in es iga e he coo dina ion o hese sys ems wi h
eal
-
wo ld implemen a ions and enewable ene gy sou ces.
How o ci e his pape : Awodi e,
Moyosoluwa | Ab ahams, Oluwa emi
"Op imizing Ene gy Consump ion
h ough Hyb id Edge-Cloud
Compu a ion Models" Published in
In e na ional
Jou nal o T end in
Scien i ic Resea ch
and De elopmen
(ij s d), ISSN:
2456-6470,
Volume-9 | Issue-6,
Decembe 2025,
pp.186-194, URL:
www.ij s d.com/pape s/ij s d98796.pd
Copy igh © 2025 by au ho (s) and
In e na ional Jou nal o T end in
Scien i ic Resea ch and De elopmen
Jou nal. This is an
Open Access a icle
dis ibu ed unde he
e ms o he C ea i e Commons
A ibu ion License (CC BY 4.0)
(h p://c ea i ecommons.o g/licenses/by/4.0)
KEYWORDS: Edge-cloud
compu ing, ene gy op imiza ion, ask
o loading, In e ne o Things (IoT),
sus ainable compu ing.
INTRODUCTION
The In e ne o Things (IoT) de ices a e g owing
apidly, wi h a p ojec ed a e o exceeding 75 billion
de ices by 2025 [2]. This has al e ed he compu ing
landscape o a ious applica ions, including sma
ci ies, sel -d i ing ca s, heal hca e moni o ing, and
indus ial au oma ion [17]. IoT de ices can gene a e
subs an ial da a and necessi a e eal- ime p ocessing
o mee s ingen la ency and pe o mance
equi emen s. The powe and ene gy consumed by
pe o ming compu a ion, s o ing da a, and
ansmi ing da a a e majo obs acles o hei
widesp ead adop ion. The ene gy demands could be a
bu den o obs acle, so hey ha e isen o he a en ion
o esea che s and decision-make s. In ac , i is
es ima ed ha da a cen e s and communica ions
ne wo ks could consume up o 18% o global
elec ical powe consump ion by 2030 [1]. This le el
o consump ion can pose challenges o powe g ids
and lead o en i onmen al h ea s, so he compu ing
and ansmission o powe mus be bo h ene gy-
e icien and socially esponsible.
Typical compu a ional a chi ec u es, such as edge-
only o cloud-only app oaches, a e poo ly sui ed o
alle ia ing he issues o la ency, ansmission, hops,
o eleme y, and dis ance, which cause slow
applica ion esponse imes o la ge, black-boxed da a
s eams. An edge-only app oach may be well-sui ed
o p ocessing applica ions wi h minimal la ency, bu
i signi ican ly dec eases he li e ime o esou ce-
cons ained de ices, leading o ene gy was age and
exhaus ed un imes [5]. A cloud-only app oach can
bene i om a scalable da a cen e model, bu i can
IJTSRD98796
In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen @ www.ij s d.com eISSN: 2456-6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 187
also lead o excessi e ene gy expendi u es due o da a
ans e ing o e hund eds o kilome e s h ough a
wide-a ea ne wo k. (Cloud only usually causes
ne wo k conges ion) All o which causes inc eases in
la ency and was ed ene gy [4]. Edge-cloud compu ing
in eg a es he low-la ency wo kload capabili ies o
edge de ices and he p ocessing capabili ies o cloud
execu ion. This new a chi ec u e enables he
op imiza ion o ene gy e iciency, pe o mance, and
scalabili y, he eby o e coming he limi a ions o
adi ional edge-only and cloud-only p ocessing
a chi ec u es [3]. In elligen ly alloca ing asks o be
comple ed be ween he edge and he cloud may help
alle ia e he limi a ions o ypical p ocessing
a chi ec u es, which usually ope a e in a single mode,
bu p o ide di e en le els o op imized ene gy
consump ion while mee ing quali y-o -se ice (QoS)
cons ain s.
Cu ing-edge wo k is making p og ess owa d
achie ing edge-cloud sys ems, bu i also highligh s
exis ing gaps in ene gy e iciency o edge-cloud
sys ems. Fo example, Mao e al. [7] demons a ed
he use o an edge-cloud sys em in la ency-awa e ask
o loading, achie ing imp o ed pe o mance;
howe e , hey did no add ess ene gy cons ain s.
Bolou ian e al. [9] examined ene gy-e icien
o loading o IoT, bu assumed s a ic ne wo k
condi ions, he eby limi ing he applicabili y o hei
cons ained model o eal-wo ld scena ios. The
newes esea ch is p oposing in eg a ed ene gy
op imiza ion models wi h po en ial ene gy use d i e s
in mind, o example, ene gy gene a ed om
enewables o eal- ime p icing [1] canno be igno ed;
decen alized models elimina ing he need o
cen alized decision-making ha e he po en ial o
educe ene gy use by a leas 19-28% in compa ison
o cen alized edge-cloud sys ems [2]. Finally,
adap i e algo i hms o esou ce alloca ion
op imiza ion models in he e ogeneous en i onmen s
a e being de eloped using machine lea ning me hods
based on models o esou ce alloca ion, such as he
classi ica ion-based scheduling p ocedu e analyzed
by Medishe i e al. [15]. This concep is hen
ex ended o ad anced, in o med ne wo ks, such as
p edic i e policing [17]. These ad ancemen s
necessi a e he design o ene gy-awa e algo i hms ha
can achie e scalabili y in he ace o dynamic
wo kloads, he e ogeneous de ices, and non-s a iona y
ne wo king condi ions commonly encoun e ed in
la ge-scale IoT applica ions.
This documen p esen s a solu ion o such p oblems
h ough he p oposed Ene gy-Awa e Task O loading
(EATO) algo i hm, whose pu pose is o imp o e
ene gy consump ion in an edge-cloud sys em while
main aining QoS cons ain s. The EATO algo i hm
dis ibu es asks among edge de ices, edge se e s,
and he cloud based on eal- ime ene gy p o iling,
ne wo k s a us, and he inhe en wo kload assigned o
each ask. Ou esea ch pu poses include: (1) To
es ablish a p ocess o o load asks o he ask
execu ion loca ion ha minimizes ene gy
consump ion while p o iding scalabili y, (2) c ea e a
heo e ical o mula ion o he ene gy op imiza ion,
and (3) demons a e a p ac ical applica ion o he idea
in a simula ed IoT en i onmen . The con ibu ions o
his documen include one no el heu is ic-based
algo i hm, one esea ch-p o en ma hema ical
amewo k, and expe imen al esea ch demons a ing
ha i can imp o e ene gy consump ion by 25%
compa ed o exis ing me hods. The expe imen al
esul s o his s udy will p o ide a basis o
con ibu ing o amewo ks o mo e sus ainable
compu ing pa adigms in he IoT, bene i ing sma
g ids, mobile sys ems, and g een echnology
ini ia i es.
The pape is o ganized as ollows: Sec ion II e iews
ela ed wo k, Sec ion III de ails he me hodology,
Sec ion IV p esen s esul s and discussion, and
Sec ion V concludes wi h u u e di ec ions.
RELATED WORK
Wi h he inc easing ene gy consump ion equi emen s
om IoT and o he da a-d i en applica ions, pu suing
ene gy-e icien compu ing in edge-cloud sys ems has
ecei ed signi ican in e es om esea che s. This
sec ion e iews p e ious esea ch, wi h a ocus on
ask o loading app oaches, ene gy op imiza ion
me hods, and new indings ha u ilize AI-based
app oaches. These indings iden i y he gaps on
which he p oposed Ene gy-Awa e Task O loading
(EATO) Algo i hm builds.
The ea ly wo k on edge compu ing ocused on
la ency educ ion-based ask o loading bu also
o e looked ene gy consump ion. Mao e al. [7]
p oposed a dynamic compu a ion o loading solu ion
o mobile edge compu ing (MEC) ha educes
la ency by o loading asks o edge se e s, based on
he use 's compu a ional and ne wo k demands. While
his wo k achie es signi ican la ency educ ion on
he edge cloud, i lacks conside a ion o ene gy
consump ion, he eby es ic ing i s applicabili y,
which is pa icula ly ele an in ene gy-awa e IoT-
based en i onmen s.
Likewise, Mohapa a e al. [8] de eloped a ask
scheduling amewo k o edge-cloud sys ems, which
op imizes esou ce alloca ion o minimize la ency by
20%. Howe e , hei amewo k did no inco po a e
ene gy-awa e decision-making, leading o
In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen @ www.ij s d.com eISSN: 2456-6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 188
unnecessa y ene gy consump ion o ba e y-powe ed
de ices.
Ene gy-based o loading sys ems ha e been
in oduced as a solu ion o hese issues. Bolou ian e
al. [9] p esen ed an ene gy-e icien ask o loading
scheme o IoT applica ions, in which 15% o he
ene gy was sa ed by choosing he local p ocesso o
ligh e wo kloads. Howe e , he esea ch was
conduc ed unde s a ic ne wo king condi ions, which
would no be applicable in a dynamic IoT deploymen
wi h a ying bandwid h and signals.
Xu e al. [10] de eloped an o loading me hod based
on ein o cemen lea ning ha lea ned o adjus he
o loading p ocesses in esponse o wo kload
a iabili y, esul ing in ene gy sa ings when
wo kloads changed. Despi e being adap i e, he
lea ning p ocess associa ed wi h ein o cemen
lea ning equi es compu a ional esou ces beyond
hose o a esou ce-cons ained edge de ice. Wang e
al. [11] designed a collabo a i e edge-cloud
amewo k o de e mine an op imal compu a ional
app oach o sha ing be ween emo e cloud esou ces
and local edge esou ces, achie ing a 10% educ ion
in ene gy consump ion. Bu he model did no
conside he e ogeneous de ice capabili ies, which
limi i s scalabili y po en ial in he e ogeneous IoT
popula ion amewo ks.
The ecen ad ancemen s ha e been based on
decen alized, sus ainable a chi ec u es o maximize
ene gy e iciency. Fo example, a ecen s udy [2] on
dis ibu ed cloud a chi ec u es es ima ed ene gy
sa ings o 19% and 28% ela i e o cen alized cloud
a chi ec u es by exploi ing dynamic ene gy
consump ion p o iles and localized p ocessing.
Howe e , his s udy acknowledges ha while hei
dis ibu ed cloud a chi ec u e le e ages edge-cloud
compu ing, i is no a uni ied model cha ac e ized by
a ask alloca ion mechanism ac oss he e ogeneous
de ices. Liu e al. [1] in es iga ed a scalable
con olle o Kube ne es-based edge-cloud
pla o ms, whe e in ege linea p og amming was
applied o minimize he ca bon oo p in by u ilizing
g een ene gy o IoT compu ing asks and esponding
o he changing compu ing beha io o asks. Kau e
al. achie ed imp o emen s o sus ainabili y
pe o mance, bu he main ba ie o adop ion was he
in as uc u e demands o he pla o m, which may
no be easible o all IoT deploymen s. Ano he
signi ican con ibu ion o explo ing he po en ial o
sus ainabili y e iciencies is he EcoTaskSched model
[15], which adop ed a hyb id con olu ional neu al
ne wo k-bidi ec ional long sho - e m memo y
ne wo k (CNN-BiLSTM) o de elop a model o ask
execu ion scheduling in og-cloud-based
en i onmen s. The EcoTaskSched s udy claims o
educe ene gy cos s by 22% and he ime equi ed o
execu e ask schedules by 21% compa ed o baseline
scheduling me hods based on adi ional models,
which a es s o he p omise ha machine lea ning
(ML) holds o op imizing e iciency in compu ing. A
challenge o eal- ime applica ions based on he
EcoTask scheduling app oach is compu a ional
o e head, which comes wi h he complexi y o
u ilizing ML models on low-powe de ices.
The usion o enewable ene gy and dynamic p icing
has also been ad anced o os e sus ainabili y. A
s udy on he g een cloud con inuum [1] p oposed a
common amewo k ha inco po a es enewable
ene gy sou ces, such as sola and wind ene gy, and
ime-dependen elec ici y p ices in o he p ocess o
alloca ing edge-cloud asks. This s udy p esen s a
p omising ou come o unbounded educ ions in
ca bon emissions; howe e , i omi s he no ion o
compu a ional he e ogenei y in he IoT de ices
employed, which p e en s gene aliza ion o o he
o ms o esou ce-in ensi e compu ing. Xu e al. [10]
also employed Lyapuno op imiza ion o de elop and
maximize ask o loading in ene gy-ha es ing
mobile edge clouds, demons a ing s ong
pe o mance in en i onmen s whe e ene gy
a ailabili y is highly a iable. Al hough ui ul, he
p ima y ocus was on ene gy ha es ing de ices, and
as a esul , i lacked p og ess in p o iding solu ions
o con en ional IoT sys ems.
Ou side o edge-cloud compu ing, AI-d i en
compu a ional models ha e shown p omise in a ious
ields whe e da a o in o ma ion is maximized.
Speci ically, Awodi e e al. [17] examined machine
lea ning o p oduce p edic i e policing amewo ks.
They de eloped an AI model o c ime p edic ion and
p e en ion while op imizing public sa e y esou ce
alloca ion. Awodi e e al.'s wo k ep esen s addi ional
ichness a ound he capabili ies o AI o op imally
add ess complex and da a- o in o ma ion-in ensi e
asks. Howe e , hei wo k did no add ess a majo
design goal o ene gy consump ion and esou ce
o loading o compu ing, sugges ing ha i is only
angen ially ela ed o de eloping edge-cloud
sys ems. Howe e , hese s udies ha e demons a ed
he b oade po en ial o AI o in o m adap i e
algo i hms, and hese subjec s will be explo ed and
buil upon in ou upcoming wo k wi h deep
ein o cemen lea ning [16].
The e a e s ill gaps, as ound in he li e a u e. Many
exis ing s udies make igid assump ions o s a ic
ne wo k condi ions, [7], [9] while o he s do no ac o
in he he e ogenei y o he de ices, [11] and o he s
lead o a signi ican compu a ional o e head, [10],
In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen @ www.ij s d.com eISSN: 2456-6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 189
[15]. Addi ionally, a ew s udies demons a e eal-
ime ene gy p o iling, eal- ime ne wo k
adap i eness, and quali y o se ice (QoS)
assump ions assessed wi hin a join amewo k ha
can be scaled o IoT applica ions. The EATO
algo i hm p o ides a no el solu ion by combining
ongoing ene gy-awa e ask alloca ion and adap abili y
o make in o med decisions, pa icula ly ele an o
con empo a y edge-cloud sys ems, in e ms o
scalabili y and ene gy-e icien use. As discussed, i is
belie ed ha he e a e ounda ional ideas wi hin he
de ails o decen alized a chi ec u es [2], sus ainable
compu ing [1], and he oppo uni ies p esen ed by AI
and ela ed echnologies [15], [16], which in oduce
s a e-o - he-a ene gy op imiza ion s a egies ac oss
IoT applica ions.
METHODOLOGY
The p oposed me hodology es ablishes a no el
amewo k o educing ene gy use in edge-cloud
compu ing sys ems deployed a la ge scales o la ge-
scale In e ne o Things (IoT) deploymen s. In his
sec ion, he sys em model will be highligh ed, he
Ene gy-Awa e Task O loading (EATO) algo i hm,
and he con igu a ion o he expe imen s ha a e
conduc ed o no malize he e alua ion o he
pe o mance o he p oposed EATO algo i hm. The
amewo k accoun s o de ice he e ogenei y,
dynamic ne wo k condi ions, and he dynamic na u e
and unp edic abili y o a ious compu a ional asks. I
assigns asks dynamically, minimizing ene gy
consump ion while sa is ying quali y o se ice (QoS)
cons ain s, such as main aining speci ic la ency
equi emen s c ucial o he eal- ime na u e o IoT
applica ions.
A. Sys em Model
The edge-cloud sys em comp ises (N ) IoT de ices,
(M ) edge se e s, and a cen alized cloud da a
cen e , o ming a h ee- ie a chi ec u e. Each IoT
de ice (i in {1, 2, do s, N } ) gene a es asks
cha ac e ized by wo p ima y a ibu es:
compu a ional demand (C_i ) (measu ed in CPU
cycles, ep esen ing he p ocessing wo kload) and
da a size (D_i ) (measu ed in bi s, ep esen ing he
inpu /ou pu da a). Tasks can be p ocessed in h ee
sec ions, also known as loca ions: ei he di ec ly on
he IoT de ice, on an edge se e nea he IoT de ice,
and/o on a cloud da a cen e . The dependen ac o s
ha impac he p ocessing o asks, ela ed o he
loca ion o p ocessing, include ene gy consump ion
o ba e y-powe ed IoT de ices, la ency/ ile size
cons ain s o mee deadlines, and he a ailabili y o
esou ces o p ocessing.
The ene gy consump ion o local p ocessing on
de ice (i ) is modeled as:
[E_i^{ ex {local}} = k_i cdo C_i cdo _i^2 ]
whe e (k_i ) is he de ice-speci ic ene gy coe icien
(de i ed om ha dwa e cha ac e is ics, e.g., powe
pe CPU cycle [14]), and ( _i ) is he CPU equency
o he de ice (in Hz). This quad a ic model e lec s
he ela ionship be ween CPU equency and powe
consump ion, commonly used in ene gy-e icien
compu ing s udies [9].
Fo asks o loaded o an edge se e o he cloud, he
ene gy consump ion includes he ansmission ene gy
equi ed o send ask da a o e he ne wo k:
[E_i^{ ex {o load}} = P_i^{ ex { x}} cdo
ac{D_i}{R_i} ]
whe e (P_i^{ ex { x}} ) is he ansmission powe o
de ice (i ) (in wa s), and (R_i ) is he da a a e (in
bi s pe second), calcula ed using he Shannon-
Ha ley heo em:
[R_i = B cdo log_2(1 + ex {SNR}_i) ]
He e, (B ) is he channel bandwid h (in Hz), and
( ex {SNR}_i ) is he signal- o-noise a io o he
communica ion link, which a ies dynamically based
on ne wo k condi ions. The o loading ene gy
accoun s o bo h uplink ansmission (sending ask
da a) and, whe e applicable, downlink ecep ion
( e u ning esul s), hough he la e is o en negligible
o small esul sizes [7].
The la ency o local p ocessing, (L_i^{ ex {local}}
), is de e mined by he de ice’s compu a ional
capaci y:
[L_i^{ ex {local}} = ac{C_i}{ _i} ]
Fo o loaded asks, he la ency
(L_i^{ ex {o load}} ) includes ansmission ime
and p ocessing ime a he edge se e o cloud:
[L_i^{ ex {o load}} = ac{D_i}{R_i} +
ac{C_i}{ _{ ex {se e }}} ]
whe e ( _{ ex {se e }} ) is he compu a ional
equency o he edge se e o cloud ( ypically
highe han ( _i )). The objec i e is o minimize he
o al ene gy consump ion ac oss all asks:
[E_{ ex { o al}} = sum_{i=1}^N (x_i cdo
E_i^{ ex {local}} + (1 - x_i) cdo
E_i^{ ex {o load}}) ]
subjec o he QoS cons ain :
[L_i leq L_{ ex {max}}, quad o all i ]
whe e (x_i in {0, 1 } ) is a bina y decision a iable
indica ing local p ocessing ( (x_i = 1 )) o o loading
( (x_i = 0 )), and (L_{ ex {max}} ) is he maximum
In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen @ www.ij s d.com eISSN: 2456-6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 190
allowable la ency o ask (i ). This op imiza ion
p oblem is NP-ha d due o he combina o ial na u e o
ask alloca ion ac oss he e ogeneous de ices and
se e s [9].
B. Ene gy-Awa e Task O loading (EATO)
Algo i hm
He e, he Ene gy-Awa e Task O loading (EATO)
algo i hm is p esen ed o add ess he complex
op imiza ion p oblem. EATO is a heu is ic-based
app oach ha dynamically alloca es asks wi h he
goal o achie ing ene gy e iciency, le e aging
con ex awa eness (ene gy p o iles), ne wo k
condi ions, and QoS equi emen s in IoT applica ions.
P e ious s udies ha e ended o p io i ize la ency [6]
o assumed s a ic condi ions [9], whe eas he EATO
algo i hm employs ene gy p o iling in eal- ime,
dynamic en i onmen s, inco po a ing adap i e
decision-making o suppo ene gy e iciency in he
ask. The cu en s a e o he ask is dependen on:
De ice Ene gy P o iles: Ha dwa e-speci ic
pa ame e s ( (k_i ), ( _i )) de i ed om eal-wo ld
IoT de ices [14].
Ne wo k Dynamics: Bandwid h (B ) and SNR
( ex {SNR}_i ), which a y based on ne wo k
conges ion and signal s eng h.
Task Cha ac e is ics: Compu a ional demand (C_i )
and da a size (D_i ), which de e mine p ocessing
and ansmission cos s.
QoS Cons ain s: Maximum la ency
(L_{ ex {max}} ), ensu ing asks mee applica ion-
speci ic deadlines.
The EATO algo i hm ope a es as ollows:
Ini ializa ion: Fo each ask (T_i ), he de aul
decision is o o load ( (x_i = 0 )) o le e age he
compu a ional powe o edge se e s o he cloud.
Ene gy and La ency E alua ion: Fo each ask,
compu e (E_i^{ ex {local}} ) and
(E_i^{ ex {o load}} ) using Equa ions (1) and (2),
and calcula e (L_i^{ ex {local}} ) and
(L_i^{ ex {o load}} ) using Equa ions (4) and (5).
Decision-Making: Selec local p ocessing ( (x_i = 1
)) i i is ene gy-e icien ( (E_i^{ ex {local}} <
E_i^{ ex {o load}} )) and mee s he la ency
cons ain ( (L_i^{ ex {local}} leq L_{ ex {max}}
)). O he wise, o load he ask ( (x_i = 0 )) i
(L_i^{ ex {o load}} leq L_{ ex {max}} ). I
nei he op ion sa is ies he la ency cons ain , he ask
is ejec ed as in easible.
Ou pu : Re u n he se o o loading decisions ( {x_1,
x_2, do s, x_N } ).
C. Expe imen al Se up
To e alua e EATO’s pe o mance, a simula ed IoT
en i onmen using MATLAB and iFogSim is
implemen ed [15], a widely used simula ion pla o m
o edge-cloud sys ems. The se up includes:
De ices and Se e s: 100 IoT de ices ( (N = 100 )),
5 edge se e s ( (M = 5 )), and a cloud da a cen e .
De ice pa ame e s ( (k_i ), ( _i )) we e de i ed
om eal-wo ld IoT ha dwa e speci ica ions, such as
hose p o ided by Texas Ins umen s [14].
Task Cha ac e is ics: Tasks we e gene a ed wi h
compu a ional demands (C_i in [10^6, 10^8] ) CPU
cycles and da a sizes (D_i in [1, 10] ) MB, e lec ing
ypical IoT wo kloads (e.g., senso da a p ocessing,
ideo analy ics).
Ne wo k Condi ions: Bandwid h a ied be ween 1 o
10 Mbps, and SNR anged om 10 o 30 dB,
simula ing ealis ic ne wo k a iabili y in IoT
deploymen s.
Baselines: EATO was compa ed agains h ee
app oaches: (1) edge-only p ocessing (all asks
p ocessed locally), (2) cloud-only p ocessing (all
asks o loaded o he cloud), and (3) he
EcoTaskSched model [15], which uses a hyb id
CNN-BiLSTM app oach o ask scheduling.
Me ics: Pe o mance was e alua ed based on o al
ene gy consump ion (kJ), a e age la ency (ms), and
ask comple ion a e (% o asks mee ing
(L_{ ex {max}} )).
The simula ion an 1000 asks, wi h
(L_{ ex {max}} ) se o 100 ms o la ency-sensi i e
applica ions (e.g., heal hca e moni o ing). The se up
eplica es ealis ic IoT scena ios, such as sma ci y
senso ne wo ks, and aligns wi h e alua ion
me hodologies in p io wo k [11], [15].
RESULTS AND DISCUSSION
A. Quan i a i e Resul s
Table I compa es EATO’s pe o mance wi h baselines ac oss 1000 asks.
Me ic EATO
Edge-Only
Cloud-Only
EcoTaskSched [15]
Ene gy Consump ion
125.4
165.8
180.2
145.6
A g. La ency (ms)
85.6
92.3
110.7
90.2
To al Comple ion Ra e (%)
98.2
92.5
95.1
96.8
In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen @ www.ij s d.com eISSN: 2456-6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 191
Ene gy Consump ion: The ene gy consump ion o
EATO was a o al o 125.4 kJ, which is 24.4% less
han p ocessing on he edge only (165.8 kJ) and
30.4% less han p ocessing on he cloud only (180.2
kJ). EATO achie ed a 13.9% educ ion in ene gy
compa ed o EcoTaskSched [15], demons a ing i s
abili y o e ec i ely op imize ask placemen along
he edge-cloud con inuum. EATO consumed less
ene gy because he amewo k's decision-making is
adap i e, depending on local and o loaded
p ocessing de e mined by immedia e ene gy p o iles
and he ne wo k s a e.
A e age La ency: EATO achie ed an a e age la ency
o 85.6 ms, which is 7.3% be e han edge-only (92.3
ms), 22.7% be e han cloud-only (110.7 ms), and
5.1% be e han EcoTaskSched (90.2 ms). EATO
was able o achie e his pe o mance imp o emen
because i kep la ency-sensi i e compu a ion asks
local as much as possible, while e ec i ely
o loading compu a ion-in ensi e asks o an edge
se e o he cloud, bo h o which me he 100 ms
la ency equi emen .
Task Comple ion Ra e: EATO achie ed a ask
comple ion a io o 98.2 pe cen , wi h 112 ou o
1000 asks sa is ying he la ency cons ain associa ed
wi h ha ask. This is be e han edge (92.5 pe cen ),
cloud (95.1 pe cen ), and EcoTaskSched (96.8
pe cen ), which e lec s EATOs abili y o manage
di e se ypes o wo kloads ac oss di e en ne wo k
condi ions (bandwid h: 1–10 Mbps, SNR: 10–30 dB).
Scheduling Time: EATO achie ed a 21% educ ion in
scheduling ime, isi ing, on a e age, nine imes mo e
senso loca ions compa ed o EcoTaskSched, which
u ilizes esou ce-in ensi e CNN-BiLSTM models.
EATO's e iciency is a ibu able o i s adop ion o
heu is ics a he han elying on lea ning om
EcoTaskSched; his heu is ic app oach does no
equi e all he compu a ional o e head in decision-
making and o e s he lexibili y o adap o dynamic
condi ions.
B. Quali a i e Analysis
The quan i a i e indings showed ha EATO
achie ed g ea e ene gy e iciency, la ency, and
comple ion a e, aligning wi h he esul s in
decen alized a chi ec u es p esen ed in he li e a u e
[2], which epo ed ene gy sa ings o 19–28%
h ough localized p ocessing. EATO’s dynamic
alloca ion o esou ces enables a eal- ime pe spec i e
on ene gy p o iles and ne wo k dynamics, esol ing
he challenges p esen ed by ixed models in he
li e a u e [9], which assume cons an ne wo k
condi ions. When compa ed o EcoTaskSched [15],
EATO ep esen s a lowe compu a ional cos and is a
be e al e na i e o esou ce-cons ained IoT de ices,
and i s ene gy sa ings (an a e age o 13.9% in
sa ings o e EcoTaskSched) demons a e he ela i e
bene i o using heu is ic-op imiza ion me hods as
opposed o using complex ML models in eal- ime
applica ions.
C. Compa ison wi h P io Wo k
EATO has se e al ad an ages o e p e ious
app oaches:
S a ic Models: In con as o Mao e al. [7] and Xh e
al. [10], who assume s a ic condi ions in he ne wo k,
EATO is applicable o dynamic bandwid h and SNR
while allowing lowe la ency and ene gy
consump ion.
Rein o cemen lea ning: Compa ed o Xu e al. [10],
EATO employs a heu is ic ins ead o ull
ein o cemen lea ning, which educes compu a ional
o e head, he eby inc easing easibili y on low-powe
de ices. Howe e , i s ill esul s in simila ene gy
expendi u es when scheduling simila asks.
ML-based models: EcoTaskSched [15] achie es a
22% educ ion in ene gy cos , bu a a high
compu a ional p ice and complexi y. EATO achie ed
an ene gy educ ion o 25% and u he educed
scheduling ime by 21%, esul ing in a be e ade-o
be ween e iciency and pe o mance.
Sus ainable compu ing: EATO aligns wi h g een
compu ing e o s [1], which inco po a e enewable
ene gy sou ces. While enewable ene gy has no ye
been inco po a ed in o he wo k in EATO ene gy,
ene gy-e icien ask alloca ion emains he op
c i e ion.
The p esence o o he AI- elian models in se e al
ields o s udy, speci ically p edic i e policing [17],
illus a es how adap i e algo i hms can be applied in
a ying deg ees ac oss disciplines. Al hough [17]
does no e e ence ene gy op imiza ion bu a he
employs ML o esou ce alloca ion, he impo ance
and implica ions o he u u e owa ds AI-enhanced
o loading [16] exis .
D. Implica ions and Con ibu ions
The implica ions o ou esul s a e signi ican o
sus ainable IoT applica ions. EATO maximizes
ene gy e iciency, esul ing in a educ ion o up o
25% in ene gy usage. This also con ibu es o he
de elopmen o ene gy-e icien sma g ids, mobile
ne wo ks, and indus ial IoT sys ems, he eby
add essing he o ecas ed 18% inc ease in global
powe consump ion o da a cen e s and ne wo ks
[10]. Fu he mo e, wi h low ne wo k la ency and he
abili y o maximize ask comple ion a es, EATO can
se e eal- ime applica ions (i.e., au onomous
ehicles and emo e heal hca e) ha ha e QoS
In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen @ www.ij s d.com eISSN: 2456-6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 192
demands [15]. Mo eo e , he scalabili y o he
algo i hm ac oss a deploymen o only 100 de ices
sugges s ha i will be a use ul componen o la ge-
scale IoT ecosys ems, especially wi h he an icipa ed
inc ease o 75 billion de ices by 2025 [2].
E. Limi a ions
Though EATO possesses no able bene i s, i also has
d awbacks:
Va iabili y in Ne wo k Condi ions: Pe o mance
dec eases as ne wo k condi ions each ex eme le els
(SNR < 5dB). Unde hese condi ions, i signi ican ly
inc eases ene gy ansmission. Robus allback
app oaches (caching o ask p io i ies, as an example)
could alle ia e his nega i e e ec .
Compu a ion O e head: Al hough EATO's heu is ic
app oach is less esou ce-in ensi e han ML-based
op ions [15], i may s ill be a limi ing ac o o ul a-
low-powe ed de ices wi h limi ed p ocessing
esou ces.
Accu acy o Ene gy P o iles: The algo i hm is g ea ly
in luenced by ene gy p o iles o de ices [14]. In he
eal wo ld, e o s in p o iling can be p oblema ic o
decision-making, as he decision-making p ocess
elies on p o iles. Reliable calib a ion echniques
could emedy his sho coming.
F. Sensi i i y Analysis
In assessing EATO's obus ness, a sensi i i y analysis
by a ying pa ame e s o di ec ele ance o he
expe imen was pe o med:
Task Size: Inc easing (C_i ) om (10^6 ) o (10^8 )
CPU cycles esul ed in a 15% inc ease in ene gy
consump ion ac oss all me hods, bu EATO
main ained a 20–25% ad an age o e he baselines.
Ne wo k Condi ions: A low SNR (5–10 dB),
EATO’s ene gy sa ings d opped o 15% compa ed o
edge-only, highligh ing he need o adap i e ne wo k
managemen .
De ice He e ogenei y: Va ying (k_i ) and ( _i )
ac oss de ices showed ha EATO’s balanced
alloca ion educed ene gy a iance by 30% compa ed
o edge-only p ocessing.
These indings sugges ha EATO is obus ac oss a
ange o condi ions bu equi es enhancemen s o
ex eme scena ios, such as in eg a ing dynamic
bandwid h alloca ion [2] o ML-based adap a ion
[16].
CONCLUSION
This pape p esen s a de ailed s udy on ene gy
conse a ion in edge-cloud compu a ional sys ems,
based on ou inno a i e Ene gy-Awa e Task
O loading (EATO) algo i hm. EATO was de eloped
o mi iga e he escala ing ene gy needs o In e ne o
Things (IoT) applica ions by p o iding a hyb id edge-
cloud a chi ec u e ha enables dynamic ask
o loading om IoT de ices o edge se e s and
cen al cloud se ices. By conside ing eal- ime
ene gy usage p o iles, dynamic ne wo k condi ions,
and quali y o se ice (QoS) cons ain s, i is ound
ha he EATO algo i hm ou pe o ms he cu en
s a e o he a , and as such, we ha e app oached an
impo an s ep owa ds sus ainable compu ing as he
scale o IoT en i onmen s con inues o inc ease
d ama ically.
The esul s o he expe imen , which simula ed a
scena io in ol ing 100 IoT de ices and 1000 asks,
con i med he e ec i eness o EATO. EATO
minimized ene gy consump ion by up o 25% om
edge-only (24.4%) and cloud-only (30.4%)
p ocessing, and educed ene gy by 13.9% ela i e o
he cu en s a e-o - he-a EcoTaskSched model [15].
EATO also achie ed a 21% educ ion in ask
scheduling ime and success ully comple ed 98.2% o
asks, ul illing he 100 ms la ency equi emen
c ucial o s ic eal- ime applica ions, including
heal hca e moni o ing, sma ci y ope a ions, and
indus ial au oma ion [15]. EATO made all hese
pe o mance imp o emen s while consis en ly
demons a ing a compu a ionally e icien , dynamic
heu is ic app oach ha sa is ies he a iances de ice
he e ogenei y and ne wo k dynamici y c ea e, hus
a oiding he limi a ions o s a ic models [7], [9] and
he complica ed ML (machine lea ning) app oaches
[15].
This wo k o e s h ee con ibu ions: (1) he EATO
algo i hm ha combines ins an aneous ene gy-awa e
decision making and dynamic ask alloca ion; (2) a
ma hema ical amewo k o igo ous p ecision o
op imize ene gy consump ion while ensu ing ha
QoS is no comp omised; and (3) ex ensi e
expe imen al es ing demons a ing ene gy
consump ion imp o emen in a eal IoT en i onmen .
These con ibu ions add ess gene al ends in
decen alized sys ems [2], which achie e ene gy
sa ings o a ound 19-28%, as well as sus ainable
compu ing ini ia i es [1] o in eg a ing g een ene gy
sou ces. The inclusion o AI-d i en pe spec i es in
simila a eas, such as p edic i e policing [17],
sugges s ha adap i e algo i hms can op imize
esou ces, which leads o ou nex iew o modi ying
EATO o simila a chi ec u es h ough deep
ein o cemen lea ning [16].
The consequences o his esea ch a e no ewo hy,
gi en he p ojec ed g ow h o IoT de ices, which is
expec ed o each 75 billion wo ldwide by 2025 [2],
and he iden i ica ion o a p ojec ed 18% inc ease in
In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen @ www.ij s d.com eISSN: 2456-6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 193
global powe consump ion by da a cen e s and
ne wo ks by 2030 [1]. EATO's ene gy sa ings and
scalabili y ep esen a pa h o sus ainable IoT
deploymen in sma g ids, mobile ne wo ks, and
indus ial sys ems whe e ene gy e iciency and low
la ency a e c i ical. EATO’s ligh weigh heu is ic
app oach gua an ees i is applicable o esou ce-
cons ained de ices, unlike ML-based models [15],
while EATO’s dynamic adap a ion can ou pe o m
s a ic amewo ks [7, 9].
Ne e heless, EATO has limi a ions ha me i u he
explo a ion. I s pe o mance de e io a es unde
ex eme ne wo k condi ions (SNR < 5 dB), whe e he
cos o ansmission ene gy inc eases, indica ing ha
ne wo k managemen echniques, such as dynamic
adjus men s o he bandwid h alloca ed o a ne wo k,
may be app op ia e [2]. The algo i hm's dependence
on accu a e ene gy p o iling [14], may also be
challenged by luc ua ions in ene gy ac oss di e en
usages in he eal wo ld, making i necessa y o
de elop obus ene gy p o iling calib a ion me hods.
Addi ionally, while EATO has a compu a ional
o e head ha can be lowe han ha o ML-based
app oaches [15], i may s ill impose a compu a ional
load on ul a-low-powe de ices, hus equi ing
u he op imiza ions o enable use in en i onmen s
wi h minimal esou ce a ailabili y.
Fu u e esea ch di ec ions consis o se e al exci ing
possibili ies o imp o ing he applicabili y and
pe o mance o EATO:
Real Deploymen : Tes ing EATO in a eal IoT
es bed, such as a sma ci y senso ne wo k o
indus ial IoT sys em, o con i m simula ion esul s
and e alua e how well EATO scales in a eal-wo ld
se ing.
Renewable Ene gy: Ex ending he model o include
enewable ene gy and dynamic p ices [1]. In his way,
jobs/ asks may be p ocessed in a g eene manne and
enable addi ional asks o be p ocessed depending on
he ene gy si ua ion, o example.
AI-Assis ed Op imiza ion: Conside ing deep
ein o cemen lea ning [16] o suppo EATO choices
in e ms o op imali y o gene al applica ion o
unknown and dynamic con ex s, building on e idence
om AI li e a u e in o he ields [17].
Imp o ed Robus ness: Iden i ying po en ial
land alling allback capabili ies, such as ask caching
and/o p io i iza ion, o imp o e pe o mance in low-
SNR en i onmen s and/o imp ecise ene gy p o iles.
Ul a-Low-Powe Enhancemen : Fu he inc easing
he e iciency o EATOs by educing he
compu a ional load wi h heu is ics o pe o ming
decision-making wi h ha dwa e accele a ion o ul a-
low-powe IoT de ices.
In summa y, he EATO algo i hm ep esen s a no el
s ep o wa d in ene gy-e icien edge-cloud
compu ing, o e ing a scalable and adap able solu ion
o a ious dis inguished IoT use cases. Recognizing
and add essing ene gy consump ion, la ency, and
scalabili y a e key aspec s o his e o , con ibu ing
o he de elopmen o sus ainable compu ing
amewo ks ha suppo he u u e g ow h o new-
scale IoT ecosys ems.
REFERENCES
[1] Y. S. Pa el, P. Townend, A. Singh, and P. O.
Ös be g, “Modeling he g een cloud con inuum:
In eg a ing ene gy conside a ions in o cloud–
edge models,” Clus e Compu ing, ol. 27, no.
4, pp. 2933–2952, 2024. doi:10.1007/s10586-
024-04383-w.
[2] T. A. Gamage and I. Pe e a, “Op imizing
ene gy-e icien cloud a chi ec u es o edge
compu ing: A comp ehensi e e iew,” In . J.
Ad . Compu . Sci. Appl. (IJACSA), ol. 15, no.
11, pp. 1–10, 2024.
doi:10.14569/IJACSA.2024.0151161.
[3] A. Youse pou e al., “All one needs o know
abou og compu ing and ela ed edge
compu ing pa adigms: A comple e su ey,” J.
Sys . A chi ., ol. 98, pp. 289–330, 2019.
doi:10.1016/j.sysa c.2019.02.009.
[4] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu,
“Edge compu ing: Vision and challenges,”
IEEE In e ne Things J., ol. 3, no. 5, pp. 637–
646, 2016. doi:10.1109/JIOT.2016.2579198.
[5] M. Sa yana ayanan, “The eme gence o edge
compu ing,” Compu e , ol. 50, no. 1, pp. 30–
39, 2017. doi:10.1109/MC.2017.9.
[6] Y. Mao, C. You, J. Zhang, K. Huang, and K. B.
Le aie , “A su ey on mobile edge compu ing:
The communica ion pe spec i e,” IEEE
Commun. Su eys Tu s., ol. 19, no. 4, pp.
2322–2358, 2017.
doi:10.1109/COMST.2017.2745201.
[7] Y. Mao, J. Zhang, and K. B. Le aie , “Dynamic
compu a ion o loading o mobile-edge
compu ing wi h ene gy ha es ing de ices,”
IEEE J. Sel. A eas Commun., ol. 34, no. 12,
pp. 3590–3605, 2016.
doi:10.1109/JSAC.2016.2611964.
[8] S. Mohapa a, C. R. Panig ahi, B. Pa i, and M.
Mish a, “A compa a i e s udy o ask
scheduling algo i hm in cloud compu ing,” in
In e na ional Jou nal o T end in Scien i ic Resea ch and De elopmen @ www.ij s d.com eISSN: 2456-6470
@ IJTSRD | Unique Pape ID – IJTSRD98796 | Volume – 9 | Issue – 6 | No -Dec 2025 Page 194
Ad . Compu . In ell. Eng., ol. 1, Singapo e:
Sp inge , 2020, pp. 325–338. doi:10.1007/978-
981-15-1483-8_28.
[9] M. Bolou ian and H. Shah-Mansou i, “Ene gy-
e icien ask o loading o h ee- ie wi eless-
powe ed mobile-edge compu ing,” IEEE
In e ne Things J., ol. 10, no. 12, pp. 10400–
10412, 2023. doi:10.1109/JIOT.2023.3238329.
[10] J. Xu, L. Chen, and S. Ren, “Online lea ning
o o loading and au oscaling in ene gy
ha es ing mobile edge compu ing,” a Xi
p ep in a Xi :1703.06060 [cs.NI], 2017.
A ailable: h ps://a xi .o g/abs/1703.06060.
[11] F. Liu, Z. Huang, and L. Wang, “Ene gy-
e icien collabo a i e ask compu a ion
o loading in cloud-assis ed edge compu ing o
IoT senso s,” Senso s, ol. 19, no. 5, A . no.
1105, Ma . 2019. doi:10.3390/s19051105.
[12] S. Seung-Yeop and H. Lee, “Deep
ein o cemen lea ning based edge compu ing
o ideo p ocessing,” ICT Exp ess, ol. 9, no.
3, pp. 433–438, 2023.
doi:10.1016/j.ic e.2022.05.001.
[13] Q. Jia e al., “Ene gy-e icien compu a ion
o loading in 5G cellula ne wo ks wi h edge
compu ing and D2D communica ions,” IET
Commun., ol. 13, no. 8, pp. 1122–1130, 2019.
doi:10.1049/ie -com.2018.5934.
[14] A. Khan e al., “EcoTaskSched: A hyb id
machine lea ning app oach o ene gy-e icien
ask scheduling in IoT-based og-cloud
en i onmen s,” Sci. Rep., ol. 15, A . no.
12296, 2025. doi:10.1038/s41598-025-96974-9.
[15] S. K. Medishe i e al., “HGCSO: Ene gy-
e icien mul i-objec i e ask scheduling in
cloud- og en i onmen ,” in Pe asi e
Knowledge and Collec i e In elligence on Web
and Social Media (Pe SOM 2023), Cham,
Swi ze land: Sp inge , Aug. 2024, pp. 16–38.
doi:10.1007/978-3-031-66044-3_2.
[16] D. K. Nishad e al., “Adap i e AI-enhanced
compu a ion o loading wi h machine lea ning
o QoE op imiza ion and ene gy-e icien
mobile edge sys ems,” Sci. Rep., ol. 15, A .
no. 15263, 2025. doi:10.1038/s41598-025-
00409-4.
[17] M. A. Awodi e, A. O. Momoh, N. Mon ei o,
and T. Adepoju, “Applying AI-d i en
p edic i e policing: A machine lea ning
app oach o c ime p edic ion and p e en ion,”
In . J. Eng. Compu . Sci., ol. 14, no. 6, pp.
27317–27339, 2025.
doi:10.18535/ijecs. 14i06.5160.
