WFQ-Based SLA-Aware Edge Applications Provisioning

Pedro Henrique Sachete Garcia

1 a

, Arthur Francisco Lorenzon

3 b

, Marcelo Caggiani Luizelli

1 c

Paulo Silas Severo de Souza

1 d

and F

abio Diniz Rossi

2 e

Federal University of Pampa, Brazil

Federal Institute Farroupilha, Brazil

Federal University of Rio Grande do Sul, Brazil

ﬂ

Keywords:

Application Performance, Edge Provisioning, Resource Allocation.

Abstract:

Provisioning delays can severely degrade the performance of real-time applications, especially in critical sec-

tors such as healthcare, smart cities, and autonomous vehicles, where fast and reliable responses are essential.

While managing data trafﬁc effectively, existing ﬂow scheduling techniques often fail to account for high-

level metrics like Service-Level Agreements, leading to suboptimal prioritization of critical applications. This

paper introduces a novel algorithm designed to optimize network bandwidth allocation for edge applications

by incorporating SLA-based metrics. Evaluations demonstrate that our proposal outperforms conventional

scheduling techniques. The results show that under varying infrastructure usage levels, our proposal consis-

tently reduces provisioning times and minimizes delay violations for edge applications.

1 INTRODUCTION

The provisioning of edge applications is currently fac-

ing a pressing challenge (Temp et al., 2023). The in-

creasing complexity and volume of connected devices

are demanding fast and efﬁcient responses. Any de-

lay in this provisioning process can compromise the

performance of applications that rely on low latency

and high availability, which are essential characteris-

tics in edge computing environments (Temp. et al.,

2023). In scenarios with high demand for process-

ing and data transfer, even a slight delay in provi-

sioning can lead to service degradation, performance

drops, interruptions, or even system failures. These

outcomes are unacceptable in critical contexts such as

in industries and smart cities. The delay in provision-

ing can be caused by infrastructure overload, lack of

resource prioritization, or poor data ﬂow management

(Souza. et al., 2022).

Flow scheduling techniques are one approach to

addressing this demand for prioritization in edge ap-

https://orcid.org/0009-0005-7487-854X

https://orcid.org/0000-0002-2412-3027

https://orcid.org/0000-0003-0537-3052

https://orcid.org/0000-0003-4945-3329

https://orcid.org/0000-0002-2450-1024

plication provisioning (Cho and Easwaran, 2020).

These techniques are designed to dynamically man-

age and allocate data ﬂows, prioritizing those that re-

quire more bandwidth and minimal latency. In edge

computing, ﬂow scheduling can organize data trafﬁc,

allowing critical applications, such as the ones men-

tioned, to gain priority access to network resources

(Tang et al., 2021). This way, it is possible to op-

timize provisioning time, minimizing delays for the

applications that need the fastest response. However,

most current ﬂow scheduling techniques have a sig-

niﬁcant limitation: they do not consider high-level

metrics, such as compliance with the service level

agreements (SLAs) of the applications. These SLAs

establish performance and availability parameters that

must be maintained to ensure the quality of the ser-

vice provided (Ramneek and Pack, 2021). Ignoring

these metrics can result in poor prioritization of ﬂows,

where critical applications may not receive the proper

attention in terms of resource allocation, even though

they theoretically need wider bandwidth or lower la-

tency.

The article introduces a tool called SLWQ

(Service-Level Weighted Queuing), which focuses on

efﬁcient network bandwidth allocation to accelerate

the transfer of container images over the network,

thereby improving the speed and efﬁciency of provi-

Garcia, P. H. S., Lorenzon, A. F., Luizelli, M. C., Severo de Souza, P. S. and Rossi, F. D.

WFQ-Based SLA-Aware Edge Applications Provisioning.

DOI: 10.5220/0013268000003950

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 15th International Conference on Cloud Computing and Services Science (CLOSER 2025), pages 159-166

ISBN: 978-989-758-747-4; ISSN: 2184-5042

159

sioning edge applications. SLWQ is an adaptation of

the Weighted Fair Queueing (WFQ) algorithm, incor-

porating the applications SLA as a weight to allocate

network resources more effectively. By prioritizing

applications with higher service requirements, SLWQ

ensures that critical applications receive more band-

width, leading to faster provisioning times. Although

WFQ was originally developed for managing network

ﬂow control, SLWQ extends this concept to resource

allocation, demonstrating its viability in contexts that

require high-level metrics to manage resource distri-

bution and guarantee compliance with service level

agreements (SLAs).

This paper is organized as follows: Section 2

presents Edge conceptualization, applications provi-

sioning ans SLA. Section 3 presents the problem def-

inition. Section 4 presents SLWQ. Section 5 presents

evaluations and results discussion. Section 6 presents

the related work. Section 7 presents conclusions and

future work.

2 BACKGROUND

Edge computing architecture is an infrastructure

model aimed at decentralizing data processing and

storage, bringing them closer to end devices and users

(Satyanarayanan et al., 2009). Instead of relying on

centralized servers in a remote data center, edge com-

puting uses distributed nodes close to data sources

to perform tasks and store information. This ap-

proach reduces latency, improves efﬁciency, and en-

ables the execution of applications that require real-

time response. Edge computing is particularly suit-

able for use cases involving the Internet of Things

(IoT) (de Souza et al., 2017), where a large number

of devices generate and consume data in nearby lo-

cations. Thus, edge architecture complements cloud

computing by keeping processing power close to the

user without the need to continuously transmit large

volumes of data to distant servers (Pitstick et al.,

2024).

Application provisioning in edge computing in-

volves several essential steps and components to en-

sure that services are made available quickly and ef-

fectively. The container-based architecture plays a

crucial role in this process (Hu et al., 2023). Contain-

ers are lightweight units that encapsulate an applica-

tion and all its dependencies, ensuring it runs consis-

tently in any environment (Xavier et al., 2013). By

using containers in edge computing, it is possible to

ensure that applications are distributed and deployed

quickly, facilitating scalability and resource manage-

ment. Container registries are repositories where con-

tainer images are stored and managed. These reg-

istries can be accessed by edge nodes, which down-

load the necessary images for application provision-

ing. Using local and distributed registries is an effec-

tive strategy to reduce provisioning latency since im-

ages are closer to the nodes that request them, avoid-

ing the transfer time from a centralized server (Temp

et al., 2024).

An aspect of successful application provisioning

in edge computing is the efﬁcient management of net-

work channels and ﬂows. Since the edge operates

with limited resources, the available bandwidth may

be scarce and needs to be used as efﬁciently as pos-

sible (Kassir et al., 2020). A well-managed network

is essential to ensure that edge computing delivers on

its promises of low latency and high efﬁciency, be-

cause network ﬂows directly inﬂuence the ability to

efﬁciently provision applications in edge computing,

allowing them to be delivered as quickly as possible

and within the parameters required by service level

agreements (Karagiannis and Papageorgiou, 2017).

3 PROBLEM DEFINITION

Figure 1 illustrates three different ﬂow control poli-

cies for computer networks, each dealing with four

data ﬂows competing for the same communication

channel. The policies shown are Equal Share, Max-

Min Fairness, and Weighted Fair Queueing, each with

its own approach to allocating the available channel

resources, resulting in different impacts on the perfor-

mance and satisfaction of competing ﬂows with vary-

ing demands.

In the Equal Share policy (Zhao and Chen, 2001),

all ﬂows receive an equal share of the total communi-

cation channel capacity, regardless of their individual

demands. Thus, each ﬂow receives 2.5 units of the

total bandwidth of 10 units. While this approach is

extremely simple and fair in terms of equality, it does

not take into account the differences in each ﬂow’s

needs. For instance, the ﬂow with a demand of 2 re-

ceives 2.5 units, which exceeds its actual need, while

other ﬂows with higher demands may not be fully sat-

isﬁed. This allocation may result in inefﬁcient re-

source usage since a ﬂow with low demand ends up

receiving more than necessary, while more demand-

ing ﬂows remain underserved.

The Max-Min Fairness policy (Min et al., 2009),

in turn, seeks to maximize the allocation for smaller

ﬂows, ensuring that each ﬂow receives the maximum

possible amount without compromising the minimum

required for other ﬂows. In the presented example, the

ﬂow with a demand of 2 receives exactly 2 units, fully

CLOSER 2025 - 15th International Conference on Cloud Computing and Services Science

160

Figure 1: Comparison of ﬂow scheduling policies.

meeting its need. The remaining ﬂows receive 2.67

units each, as, after satisfying the demand of ﬂow 2,

the rest of the bandwidth is distributed equally among

the remaining ﬂows. While this approach is fairer

by considering the speciﬁc needs of the ﬂows, it still

does not take into account differentiation of priorities

among ﬂows. Moreover, the performance of larger

ﬂows is impacted because the policy focuses on pro-

tecting smaller ﬂows, even if this results in limiting

ﬂows with higher demands.

The third policy presented is Weighted Fair

Queueing (Al-Sawaai et al., 2010), which uses dif-

ferent weights for each ﬂow, reﬂecting their rela-

tive demand. In this example, ﬂows are weighted

according to their demands and priorities: the ﬂow

with a demand of 3 receives more resources (3.63

units), followed by the ﬂow with a demand of 4 (3.45

units), then the ﬂow with a demand of 5 (2.58 units),

and lastly the ﬂow with a demand of 2 (0.34 units).

This approach tries to provide proportional treatment

to ﬂows based on their demands and priorities and

proves more efﬁcient from a bandwidth utilization

perspective, as it considers the speciﬁc needs of each

ﬂow. However, like the previous policies, Weighted

Fair Queueing ignores the importance of differentiat-

ing ﬂows based on criteria such as the criticality of the

applications or the context in which they are running.

Thus, it may not be the best approach for applications

that need faster provisioning or have speciﬁc quality

of service requirements.

Although these policies are effective in traditional

network environments, they are not suitable for edge

computing environments, where it is necessary to

consider factors beyond equitable or proportional re-

source distribution. In an edge computing scenario,

it is essential to prioritize certain applications, such

as critical ones that need to be provisioned quickly to

ensure a good user experience or to meet speciﬁc op-

erational requirements. Additionally, the size of the

data being transferred, such as large images, must be

taken into account, especially in channels with lim-

ited bandwidth, where an allocation based solely on

demand may not meet the real needs of more sensi-

tive applications.

4 SLWQ: SERVICE-LEVEL

WEIGHTED QUEUING

This work proposes an alternative approach to

network ﬂow management, called ”Service-Level

Weighted Queueing” (SLWQ), which expands upon

the concept of the Weighted Fair Queueing (WFQ)

technique. While WFQ uses weights proportional to

the demands of the ﬂows to determine bandwidth dis-

tribution, SLWQ incorporates weights derived from

high-level metrics, such as application service level

agreements (SLA). This approach allows allocations

to be adjusted not only to the ﬂows’ demands but also

to the speciﬁc requirements of each application, align-

ing resource allocation with the characteristics and

needs of each ﬂow.

Figure 2: Service-Level Weighted Queuing policy.

In Figure 2, we can see four distinct application

ﬂows, each with a speciﬁc demand and SLA. The

SLWQ policy considers both the application demand

and SLA to determine resource allocation. For in-

stance, the application with demand 2 and High SLA

receives an allocation of 2.0 units (100% of the re-

quested demand), while the application with demand

3 and Medium SLA receives 3.0 units (also 100%

of the requested demand). The remainder is divided

among applications with low SLA. This discrepancy

occurs because SLA plays a crucial role in determin-

ing the weight assigned to each application, directly

impacting how much bandwidth each one receives.

Unlike traditional WFQ, which distributes resources

based on proportional demands, SLWQ prioritizes ap-

plication ﬂows with higher levels of importance, as

WFQ-Based SLA-Aware Edge Applications Provisioning

161

indicated by the SLA metric, even if it means par-

tially sacriﬁcing the allocation of lower-SLA applica-

tion ﬂows.

In environments where multiple services compete

for the same resources, it is essential for network

management to recognize the difference in impor-

tance between these services. SLWQ allows more

sensitive applications to be prioritized, while less crit-

ical applications can be delayed or receive fewer re-

sources as long as this does not signiﬁcantly com-

promise their operation. Also, SLWQ aims to opti-

mize the use of network resources in situations where

application demands and requirements are heteroge-

neous and often conﬂicting. In edge computing en-

vironments, where resources are limited and rapid re-

sponse is essential, the use of SLWQ can ensure that

critical applications receive the necessary treatment

to meet their SLA requirements, while maximizing

overall network efﬁciency and maintaining satisfac-

tory performance for all applications involved.

5 EVALUATION AND

DISCUSSION

In this section, we evaluate the performance of the

SLWQ (Service-Level Weighted Queuing) algorithm

compared to other network provisioning policies by

analyzing the results of provisioning time (as seen

in Figure 3) and delay violations (illustrated in Fig-

ure 4). Our evaluation focuses on three distinct lev-

els of prioritization: low, medium, and high prior-

ity applications. As expected, SLWQ effectively pri-

oritizes high-priority applications over medium and

low-priority ones, demonstrating signiﬁcant improve-

ments over traditional queuing policies.

5.1 SLWQ

The Service-Level Weighted Queueing (SLWQ) pre-

sented in Algorithm 1 is an approach designed to

manage bandwidth allocation efﬁciently, driven by

service levels (SLA), and catering to the speciﬁc de-

mands of each ﬂow in a network. Table 1 consists

of notations used in the Algorithm. The core of

SLWQ lies in its ability to prioritize ﬂows with higher

criticality, as speciﬁed in the SLAs, while adjusting

the allocation according to each ﬂow’s demand re-

quirements. This process is achieved by combining

weights based on SLA and relative demands, ensur-

ing that the most critical ﬂows receive greater prior-

ity and a proportionally more signiﬁcant share of the

available bandwidth.

Table 1: SLWQ Notations.

Symbol Description

cap Total available bandwidth

dem Array of demands for each ﬂow

wgt Array of priority weights for each ﬂow

exp Exponent applied to each demand

alloc Array of allocated bandwidths

valD List of valid demands

totW Total weight for allocation

capRem Remaining capacity

The algorithm’s input consists of several essential

parameters. First, it takes the total available band-

width, which represents the maximum resource that

can be distributed among ﬂows. Next, it uses an ar-

ray of demands, where each position represents the

speciﬁc bandwidth demand of a ﬂow. There is also

an array of weights, where each weight is associated

with the priority of the corresponding ﬂow, indicating

the relative importance of the ﬂow based on its SLA.

Additionally, the algorithm uses a parameter called

“exp,” an exponent applied to each demand to cre-

ate a weighted distribution; this conﬁguration allows

for adjusting the allocation of resources according to

each demand and priority proﬁle.

Algorithm 1: Service-Level Weighted Queueing (SLWQ).

1: Input:

2: cap → Total available bandwidth

3: dem[] → Array of demands for each ﬂow

4: wgt[] → Array of priority weights for each

ﬂow

5: exp → Exponent for weighted allocation

6: Output:

7: alloc[] → Array of allocated bandwidths for

each ﬂow

8: Initialize alloc as a zero array of the same length

as dem

9: Deﬁne valD as the list of tuples (d, w, i) where

each d > 0, with w as the corresponding weight

and i as the index

10: Set totW ←

∑

(d,w,i)∈valD

exp

) × w

11: Initialize capRem ← cap

12: for each (d, w, i) in valD do

13: Compute alloc ← min



capRem×(d

exp

×w)

totW

, d



14: Update alloc[i] ← alloc[i] + alloc

15: Update capRem ← capRem − alloc

16: Update totW ← totW − (d

exp

× w)

17: if totW ≤ 0 or capRem < 0 then

18: break

19: end if

20: end for

21: return alloc

CLOSER 2025 - 15th International Conference on Cloud Computing and Services Science

162

(a) Infrastructure usage: 25%. (b) Infrastructure usage: 50%. (c) Infrastructure usage: 75%.

Figure 3: Provisioning Time: Number of violations of the SLA established as acceptable for the migration or provisioning of

each application. When this threshold is crossed, it means that users have moved to a location less close to where the service

is provisioned, and it means that a new service must be provisioned on another edge node that is closer to such users.

(a) Infrastructure usage: 25%. (b) Infrastructure usage: 50%. (c) Infrastructure usage: 75%.

Figure 4: SLA Delay Violations: Number of violations of the acceptable application delay threshold. When this threshold is

crossed, it means that the service is too far from the registry located in the core of the cloud.

The SLWQ algorithm begins by initializing an

empty bandwidth allocation array and creating a list

of valid demands, which includes each ﬂow’s de-

mand, weight, and index, ﬁltering out ﬂows with no

demand. It calculates the total weight as the sum

of each demand raised to an exponent multiplied by

its weight. The remaining bandwidth capacity is ini-

tialized to the total capacity. In the main loop, the

algorithm iterates through each ﬂow, calculating the

bandwidth allocation as a proportion of the remaining

capacity based on the ﬂow’s relative weight and de-

mand. To prevent over-allocation, the allocated band-

width is capped at the ﬂow’s demand. The allocated

bandwidth is stored in the array, and the remaining ca-

pacity and total weight are updated accordingly. The

loop ends when either all demands are met or the

bandwidth is fully allocated. The algorithm then re-

turns the ﬁnal array of allocated bandwidth values, en-

suring prioritized distribution based on demands and

SLAs, while respecting total bandwidth limits and in-

dividual ﬂow demands.

5.2 Provisioning Time Evaluation

Figure 3 presents the provisioning time across three

levels of infrastructure usage: 25%, 50%, and 75%.

Across all tests, SLWQ consistently provides the

fastest provisioning time for high-priority applica-

tions, outperforming Equal Share, Max-Min Fairness,

and even Weighted Fair Queuing (WFQ). This supe-

rior performance can be attributed to SLWQ’s abil-

ity to dynamically adjust resource allocation based on

both the priority and the demand of applications.

At 25% infrastructure usage, all policies perform

relatively well due to the availability of ample re-

sources. However, even under these low-stress con-

ditions, SLWQ demonstrates its advantage by allocat-

ing a higher proportion of bandwidth to high-priority

applications, ensuring their provisioning times remain

minimized. The differences between SLWQ and other

WFQ-Based SLA-Aware Edge Applications Provisioning

163

Table 2: Comparison between SLWQ and Evaluated Scheduling Policies.

Feature/Criteria SLWQ WFQ Max-Min Fairness Equal Share

Priority-based Resource Allocation Yes Partial No No

Latency Minimization High Medium Low Low

Bandwidth Utilization Efﬁciency High Medium Low Low

Handling of Large Data Volumes Efﬁcient Moderate Poor Poor

Real-time Application Support Strong Moderate Weak Weak

Adaptation to Infrastructure Load Yes No No No

SLA Support Yes No No No

policies become more pronounced as infrastructure

usage increases to 50% and 75%. Under higher loads,

the competing policies—particularly Equal Share and

Max-Min Fairness—fail to adequately prioritize high-

priority applications, resulting in longer provisioning

times for these critical services. In contrast, SLWQ

maintains a strategic allocation of resources, ensur-

ing that the most important applications continue to

receive the necessary bandwidth, thus minimizing de-

lays.

This advantage becomes even more critical un-

der the 75% infrastructure usage scenario, where re-

source contention is at its peak. SLWQ’s prioriti-

zation mechanism ensures that the high-priority ap-

plications, which are often latency-sensitive, experi-

ence the smallest provisioning delays. In compari-

son, WFQ, while better than Equal Share and Max-

Min Fairness, still cannot adapt as effectively to high-

priority demands since it only accounts for demand

and not priority, which is the key differentiator in

SLWQ’s design.

5.3 Delay Violations Evaluation

Figure 4 evaluates delay violations across the same

levels of infrastructure usage. Delay violations occur

when an application is not provisioned within the ex-

pected time frame, potentially leading to performance

degradation. Similar to the provisioning time results,

SLWQ exhibits a clear advantage, particularly in re-

ducing delay violations for high-priority applications.

At 25% infrastructure usage, delay violations

are minimal for all policies, as resources are abun-

dant. However, as usage climbs to 50% and 75%,

SLWQ maintains a lower incidence of delay viola-

tions compared to the other policies. For high-priority

applications, SLWQ nearly eliminates delay viola-

tions, a critical achievement for real-time or mission-

critical services, such as healthcare monitoring or au-

tonomous vehicle data processing.

The failure of Equal Share and Max-Min Fair-

ness to appropriately prioritize high-priority trafﬁc re-

sults in signiﬁcant delay violations, particularly un-

der heavy infrastructure load. These policies, which

either distribute resources equally or favor smaller

demands, cannot distinguish the urgency of high-

priority applications, leading to unacceptable delays.

Although WFQ performs better by allocating more

bandwidth to larger demands, it still results in a sub-

stantial number of delay violations under high load,

as it lacks the application-SLA focus that SLWQ im-

plements.

5.4 Discussion

The key strength of SLWQ lies in its ability to bal-

ance the competing demands of multiple applications

while ensuring that the highest-priority ones receive

preferential treatment. In contrast to Equal Share,

which treats all applications equally regardless of

their importance, and Max-Min Fairness, which pri-

oritizes smaller ﬂows, SLWQ intelligently adjusts its

resource allocation to reﬂect both the application’s de-

mand and its criticality. This results in signiﬁcantly

better provisioning times and fewer delay violations

for high-priority applications, especially in scenarios

with limited resources.

WFQ, while closer to SLWQ in terms of account-

ing for the differing demands of applications, still

falls short in environments where application prior-

ity must be considered. WFQ allocates bandwidth

proportionally based on demand but does not fac-

tor in the criticality of each application. This lim-

itation is particularly evident under high infrastruc-

ture usage, where SLWQ’s additional consideration

of service-level agreements (SLAs) allows it to out-

perform WFQ.

SLWQ’s ability to incorporate high-level met-

rics such as application priority into the queuing

mechanism ensures that the most important ser-

vices—those with stringent latency and performance

requirements—are given priority over less critical ser-

vices. This approach maximizes overall network efﬁ-

ciency and service quality, particularly in edge com-

puting environments where fast provisioning is often

essential for maintaining the functionality of real-time

applications.

The results demonstrate that SLWQ provides a

robust and effective solution for managing network

provisioning in edge computing environments. Its

ability to adapt to varying levels of infrastructure us-

age while maintaining a focus on application priority

makes it particularly suited to high-demand, resource-

constrained environments. As edge computing con-

CLOSER 2025 - 15th International Conference on Cloud Computing and Services Science

164

Table 3: Comparison between Related Work and SLWQ.

Feature/Work Shinohara et al. (1997) Jiang et al. (2000) Jutila (2016) Chen et al. (2021) SLWQ (2024)

Type of System Edge modules Cellular Networks IoT Edge Router Cooperative Games Edge Applications

Algorithm WFQ WFQ Fuzzy Weighted Queuing Nucleolus Computation SLA-based WFQ

QoS Guarantees Yes Yes Yes No Yes

Service Classes Multiple trafﬁc classes Multiple service classes Dynamic adaptation Coalition fairness High-level SLA priorities

Priority-based Allocation Partial Partial Yes No Yes

Context Awareness No No Yes Yes (Cooperative) Yes (SLA-driven)

Handling of Bursty Trafﬁc No Yes Yes No Yes

Dynamic Bandwidth Adjustment No No Yes No Yes

tinues to evolve, the need for intelligent resource allo-

cation mechanisms like SLWQ will become increas-

ingly important. Table 1 summarizes the differences

between SLWQ and other ﬂow schedulers, in several

performance evaluation metrics.

6 RELATED WORK

Weighted fair queueing (WFQ) is a scheduling algo-

rithm widely studied in network trafﬁc management.

(Shinohara et al., 1997) proposed a growable switch

architecture with multiple modules, including Edge

modules with large buffers that utilize a weighted

fair queueing scheduler for open-loop trafﬁc. This

approach allows for allocating equivalent bandwidth

based on call admission control, ensuring quality of

service (QoS) guarantees for different trafﬁc classes.

(Jiang et al., 2000) evaluated the performance of the

weighted fair queueing algorithm in providing mul-

tiple service classes for bursty data trafﬁc in cellular

networks, similar to an Edge system. The study fo-

cused on packet scheduling to achieve fair bandwidth

sharing among different types of trafﬁc, highlighting

the importance of efﬁcient queueing mechanisms in

wireless communication systems.

Furthermore, (Jutila, 2016) discussed the impor-

tance of edge computing in managing network re-

sources, trafﬁc prioritizations, and security in the In-

ternet of Things (IoT) context. The author introduced

adaptive edge computing solutions, including fuzzy

weighted queueing (FWQ), to monitor and react to

changes in the quality of service within heterogeneous

networks, emphasizing the role of edge computing

in enhancing network performance. In Cooperative

games, (Chen et al., 2021) explored the computation

of the nucleolus in various game scenarios, includ-

ing fractional edge cover games on weighted graphs.

The study focused on ﬁnding fair divisions of pay-

offs among coalition members, highlighting the sig-

niﬁcance of fair resource allocation in cooperative set-

tings.

Table 3 summarizes all related works and com-

pares them with SLWQ, comparing some metrics. Al-

tough the literature review showcases the signiﬁcance

of weighted fair queueing in optimizing network traf-

ﬁc management, ensuring QoS guarantees, and facil-

itating fair resource allocation in diverse communi-

cation systems, none of the literature considers high-

level metrics such as those service-level agreements

aimed at ensuring quality of service.

7 CONCLUSION AND FUTURE

WORK

The problem addressed in this work revolves around

the inadequacies of traditional ﬂow scheduling tech-

niques in edge computing environments. These tech-

niques, including Equal Share, Max-Min Fairness,

and Weighted Fair Queueing (WFQ), either fail to ac-

count for the speciﬁc demands of applications or do

not prioritize them according to their criticality and

Service-Level Agreements (SLAs). In edge comput-

ing environments where resources are constrained and

multiple applications are competing for bandwidth,

these approaches often result in inefﬁciencies such as

increased provisioning times and higher rates of delay

violations, particularly for high-priority applications.

In this work, we proposed the Service-Level

Weighted Queuing (SLWQ) algorithm, a novel ap-

proach designed to optimize network bandwidth al-

location for edge computing. The results of our

evaluations highlight the advantages of SLWQ over

traditional policies. In our tests, which involved

three distinct levels of application SLA (low, medium,

and high), SLWQ consistently outperformed other

policies, especially under high infrastructure usage.

When provisioning time was measured (as shown in

Figure 3), SLWQ demonstrated a signiﬁcant reduc-

tion in the time required to provision high-SLA appli-

cations, particularly when compared to Equal Share,

Max-Min Fairness, and WFQ. Additionally, SLWQ

proved to be highly effective in minimizing delay vi-

olations, as depicted in Figure 4. Delay violations oc-

cur when an application fails to be provisioned within

the expected time, which can lead to serious conse-

quences in time-sensitive applications.

Future work could explore the integration of addi-

tional factors, such as energy efﬁciency or predictive

analytics, into SLWQ’s prioritization framework.

WFQ-Based SLA-Aware Edge Applications Provisioning

165

ACKNOWLEDGEMENT

This study was ﬁnanced in part by the Coordenac¸

de Aperfeic¸oamento de Pessoal de N

ıvel Superior –

Brasil (CAPES) – Finance Code 001.

REFERENCES

Al-Sawaai, A., Awan, I. U., and Fretwell, R. (2010). Per-

formance of weighted fair queuing system with multi-

class jobs. In 2010 24th IEEE International Confer-

ence on Advanced Information Networking and Appli-

cations, pages 50–57.

Chen, N., Lu, P., and Zhang, H. (2021). Computing the

nucleolus of matching, cover and clique games. Pro-

ceedings of the AAAI Conference on Artiﬁcial Intelli-

gence, 26(1):1319–1325.

Cho, H. and Easwaran, A. (2020). Flow network models for

online scheduling real-time tasks on multiprocessors.

IEEE Access, 8:172136–172151.

de Souza, P. S. S., dos Santos Marques, W., Rossi, F. D.,

da Cunha Rodrigues, G., and Calheiros, R. N. (2017).

Performance and accuracy trade-off analysis of tech-

niques for anomaly detection in iot sensors. In 2017

International Conference on Information Networking

(ICOIN), pages 486–491.

Hu, S., Shi, W., and Li, G. (2023). Cec: A containerized

edge computing framework for dynamic resource pro-

visioning. IEEE Transactions on Mobile Computing,

22(7):3840–3854.

Jiang, Z., Chang, L. F., and Shankaranarayanan, N. (2000).

Providing multiple service classes for bursty data traf-

ﬁc in cellular networks. In Proceedings IEEE IN-

FOCOM 2000. Conference on Computer Communi-

cations. Nineteenth Annual Joint Conference of the

IEEE Computer and Communications Societies (Cat.

No.00CH37064), volume 3, pages 1087–1096 vol.3.

Jutila, M. (2016). An adaptive edge router enabling in-

ternet of things. IEEE Internet of Things Journal,

3(6):1061–1069.

Karagiannis, V. and Papageorgiou, A. (2017). Network-

integrated edge computing orchestrator for application

placement. In 2017 13th International Conference on

Network and Service Management (CNSM), pages 1–

Kassir, S., Veciana, G. d., Wang, N., Wang, X., and

Palacharla, P. (2020). Service placement for real-time

applications: Rate-adaptation and load-balancing at

the network edge. In 2020 7th IEEE International

Conference on Cyber Security and Cloud Computing

(CSCloud)/2020 6th IEEE International Conference

on Edge Computing and Scalable Cloud (EdgeCom),

pages 207–215.

Min, Z., chunming, W., Ming, J., and Jing, Y. (2009). A

resource management algorithm based on multi-path

max-min fairness. In 2009 First International Confer-

ence on Future Information Networks, pages 76–80.

Pitstick, K., Novakouski, M., Lewis, G. A., and Ozkaya, I.

(2024). Deﬁning a reference architecture for edge sys-

tems in highly-uncertain environments. In 2024 IEEE

21st International Conference on Software Architec-

ture Companion (ICSA-C), pages 356–361.

Ramneek and Pack, S. (2021). A credible service level

agreement enforcement framework for 5g edge. In

IEEE INFOCOM 2021 - IEEE Conference on Com-

puter Communications Workshops (INFOCOM WK-

SHPS), pages 1–2.

Satyanarayanan, M., Bahl, P., Caceres, R., and Davies, N.

(2009). The case for vm-based cloudlets in mobile

computing. IEEE Pervasive Computing, 8(4):14–23.

Shinohara, M., Fan, R., Mark, B., Ramamurthy, G., Suzuki,

H., and Yamada, K. (1997). Multiclass large scale atm

switch with qos guarantee. In Proceedings of ICC’97

- International Conference on Communications, vol-

ume 1, pages 453–461 vol.1.

Souza., P.,

Angelo Crestani., Rubin., F., Ferreto., T., and

Rossi., F. (2022). Latency-aware privacy-preserving

service migration in federated edges. In Proceedings

of the 12th International Conference on Cloud Com-

puting and Services Science - CLOSER, pages 288–

295. INSTICC, SciTePress.

Tang, H., Wang, Y., and Chen, X. (2021). Network-ﬂow al-

gorithms for virtual service placements in mobile edge

networks. In 2021 IEEE 6th International Conference

on Cloud Computing and Big Data Analytics (ICC-

CBDA), pages 364–368.

Temp., D., Capeletti., I., Goes de Castro., A., Silas Severo

de Souza., P., Lorenzon., A., Luizelli., M., and Rossi.,

F. (2023). Latency-aware cost-efﬁcient provisioning

of composite applications in multi-provider clouds. In

Proceedings of the 13th International Conference on

Cloud Computing and Services Science - CLOSER,

pages 297–305. INSTICC, SciTePress.

Temp, D. C., da Costa, A. A. F., Vieira, A. N. C., Oribes,

E. S., Lopes, I. M., de Souza, P. S. S., Luizelli, M. C.,

Lorenzon, A. F., and Rossi, F. D. (2024). Maper:

mobility-aware energy-efﬁcient container registry mi-

grations for edge computing infrastructures. The Jour-

nal of Supercomputing, 81(1):15.

Temp, D. C., de Souza, P. S. S., Lorenzon, A. F., Luizelli,

M. C., and Rossi, F. D. (2023). Mobility-aware reg-

istry migration for containerized applications on edge

computing infrastructures. Journal of Network and

Computer Applications, 217:103676.

Xavier, M. G., Neves, M. V., Rossi, F. D., Ferreto, T. C.,

Lange, T., and De Rose, C. A. F. (2013). Performance

evaluation of container-based virtualization for high

performance computing environments. In 2013 21st

Euromicro International Conference on Parallel, Dis-

tributed, and Network-Based Processing, pages 233–

240.

Zhao, Y. and Chen, C. (2001). The algorithm and model

of trash: A scheme enforcing equal share. In Pro-

ceedings 2001 International Conference on Computer

Networks and Mobile Computing, pages 425–432.

CLOSER 2025 - 15th International Conference on Cloud Computing and Services Science

166