Over the last few decades, information technologies in the field of computing systems and software architectures have evolved to provide companies with robust solutions to run their business logic. The transition from monolithic applications to Service-Oriented Architecture (SOA) and micro-services represents a fundamental shift in the way companies design, build, and deploy their software. This evolution has been driven by the need for greater scalability, flexibility, and agility in response to the dynamic and demanding landscape of modern application requirements. As software development continues to evolve, the incorporation of intelligent agents has become widespread to provide smartness within distributed system as well as to design and manage complex scenarios, such as the device-edgecloud continuum and IoT ecosystems. This paper explores the role of intelligent agents within SOA and micro-services architectures, highlighting the spectrum of benefits and limitations they bring to the development life cycle. We also discuss challenges common to both architectures and identify solutions that each architecture provides with respect to the other. Finally, to overcome some of these challenges, a new approach, stemmed in the context of the EU funded MLSysOps Project, is presented; the proposed approach utilizes agents and Machine Learning (ML) as-a-service to provide inbound and outbound intelligence in the system.
In recent decades, the advancement of technology and software architectures has increased
the complexity of computing environments to a degree that exceeds the manageable limits for
human system administrators. Initially, with monolithic applications, companies consolidated
data, databases, and business logic on the same host by running their applications as a single,
unified entity. In this approach, all components, including the user interface, business logic,
and database access, are tightly integrated into a unified code-base that encompassed various
services [
Service Oriented Architecture (SOA) was proposed to solve this challenges and also to
provide integration between solutions for distributed services across diferent organizations and
operational systems aiming for the reusability of a service by multiple end user applications. It
relies on simple Remote Procedure Calls (RPC), such as SOAP (Simple Object Access Protocol),
commonly used for communication. Since the advent of this software architecture,
computing environments have evolved into open and distributed systems. The typical enterprise
environment is now a complex, heterogeneous, multi-vendor landscape, posing challenges in
configuration, maintenance, and troubleshooting [
Micro-services architecture emerged as the next evolutionary step beyond SOA, with
lightweight communication channels, prioritizing simplicity in inter-service interactions. This
architectural approach directs intelligence and control to individual service endpoints, thus
enhancing the decoupling between services, addressing issues of scalability and adaptability
in the face of rapidly changing technological landscapes. In this intricate scenario of
technological progress, the trajectory from monolithic applications to SOA and, subsequently, to
micro-services architecture reflects an ongoing quest for solutions that aligns with the dynamic
nature of contemporary computing environments [
Until this end, agent-based computing (ABC) is one of the well-known paradigms that proposes an improved structuring mechanism for distributed applications. ABC is characterized by autonomous, intelligent entities capable of migrating across distributed systems; it ofers a promising avenue for addressing the escalating complexities in modern computing. In conjunction with SOA and micro-services, ABC contribute to a holistic approach that transcends traditional boundaries. The seamless integration of software agents with SOA enables the creation of agile, distributed services that can be leveraged across diverse organizations and operational systems. Moreover, when juxtaposed with micro-services architecture, ABC contributes to the ongoing quest for enhanced scalability, adaptability, and intelligence at the service endpoints. This triad of agents, SOA, and micro-services represents a forward-looking strategy, ofering a robust response to the intricate challenges posed by today’s multifaceted computing landscape.
Various studies in the field have undertaken the task of comparing mobile agents with
SOA and micro-service architectures, attempting to discern their individual strengths and
weaknesses [
The remainder of the paper is structured as follows: Section 2 describes the background, providing an introduction to the key features of the technologies and paradigm involved. Section 3 ofers an overview of the state of the art, including an analysis of the benefits and limitations of utilizing agents within both SOA and micro services architectures. The MLSysOps approach and its possible contributions to the open limitations is presented in Section 4. Final remarks conclude the paper.
This section introduces key concepts related to SOA and micro-services architectures.
SOA emerged as a response to challenges posed by large monolithic applications, aiming to enhance service re-usability across multiple end-user applications and representing an evolution from the conventional Client-Server model in distributed systems.
Web
Mobile
App Messaging Security
ENTERPRISE SERVICE BUS (ESB)
Transport
Transform Srv
Srv
Data Base
SOA promotes loosely coupled, reusable, and dynamically assembled services, providing
explicit boundaries between autonomous services on diferent servers to meet application
requirements and address the limitations of monolithic applications. A common scenario
involves clients requesting services through various user interfaces, with the Enterprise Service
Bus (ESB) handling the requests. The ESB, designed to seamlessly connect services directly
with end-users and render the physical network transparent to applications, incorporates some
service functions to link widely located services with the end user, as illustrated in Figure
1. It translates end-user requests into suitable message types for each service, fostering their
re-usability across diferent organizations for numerous customers [
In the software industry, SOA has gained substantial popularity for developing distributed enterprise-wide applications, emphasizing software re-usability and integrity in diverse environments through open standards. Companies often leverage SOA by discovering and combining internet-available services. Alternatively, some organizations prefer internal control, creating new services tailored to their specific environments with quality-attribute properties.
SOA facilitates integration solutions for distributed services across various organizations and operational systems, relying on straightforward RPC like SOAP for communication. It emphasizes the use of services across the web, aligning with the trend of distributed and interconnected systems.
In today’s digital landscape, users expect a dynamic user experience across diverse devices,
requiring frequent service updates. To meet this demand, the architectural trend emphasizes
breaking applications into smaller, independent services. This approach is often compared to
SOA, critiquing its centralized integration approach. Micro-services are viewed as an alternative,
diverging from SOA by eliminating a service bus, focusing on simple communication channels,
and enhancing service decoupling [
UI uS
uS uS uS
uS Data Base
Data Base
Data Base
Each microservice, capable of at least one functional operation, promotes independence
and reusability. Breaking databases into smaller ones benefits specific customer types,
distributing the load and allowing shared databases to operate as services. Despite numerous
benefits like technology diversity and scalability, microservices pose challenges in distributed
settings, complicating tasks like service discovery, security and resource management, as well
as communication optimization. Microservices architecture, characterized by small,
independent services running in separate processes, originated with the goal of deploying software
parts independently. These autonomous components isolate fine-grained business capabilities,
communicating via standardized interfaces and lightweight protocols. While ofering benefits
in scalability and productivity, microservices require careful consideration due to distributed
complexities[
Understanding microservices in practice is crucial, revealing motivating advantages and critical challenges. This insight support decision-making for system migration or developing applications under this architecture. Overall, the approach enhances software maintainability but demands thoughtful consideration of its challenges.
Agent-based Computing (ABC) has been recognized as a comprehensive and efective enabler for
cooperating within decentralized, dynamic, and open IoT ecosystems [
To delve further into this efectiveness, it is essential to examine the combination of agents with both SOA and microservices. Such explorations contribute to a deeper understanding of how these paradigms can be combined to optimize various aspects of system architecture and performance.
The following are some of the benefits found in the literature review of using agents with
both architectures:
• Agents in both SOA and microservices exhibit increased flexibility, autonomy, and
collaboration capabilities. In SOA, agents excel in making dynamic run time changes
based on internal rules, enhancing robustness. In microservices, agents enable adaptive
responses to change environments, contributing to improved scalability and flexibility in
system design. They eficiently distribute workloads and handle complex tasks, enhancing
overall performance and resource utilization [
On the other hand some limitations found of combining agents and SOA are grouped as
follows:
• Resource-intensive agent operations in diverse architectures, including SOA, present
challenges to overall system performance and scalability. The computational
demands and memory requirements of agents, along with potential coordination overhead
as the number of agents increases, can impact scalability. Coordinating agent behavior
within any architecture, particularly in expansive systems, poses inherent challenges.
Additionally, the training and deployment of agents in various contexts demand
substantial computational resources and processing power, potentially leading to scalability
issues [
The following limitations are also presented in the context of using agent and micro service
architectures:
• Implementing self-organization and emergence in several systems through MAS and
SOA introduces concerns due to potential risks of unpredictability and disruptions.
Security is a critical consideration, particularly when agents have access to sensitive data
and interact with external services, as their autonomous behavior can become vulnerable
to malicious attacks or unauthorized access. Addressing security in agent-based systems
within SOA necessitates robust measures, including trusted hosts, secure communication
protocols, and secure agent profiles to safeguard sensitive information. The autonomous
nature of agents raises challenges in ensuring security and privacy, making the design and
implementation of protective measures complex, especially in microservices environments
with distributed data exchange. [
Scalability and resource usage Distributed procesing and interoperability Limited reusability
Standarization Limitations of using agents with SOA and microservices
Security Performance and
overhead Complexity and specialized skills
Benefits of using agents and microservices
Resource Management and
optimization Flexibility, autonomy and adaptability Integration and colaboration Benefits of using agents with SOA and microservice Quality and service level management
Learning and proactive support Communication
Integration
management aspects, necessitating specialized skills and expertise. Factors such as
persistence management, coordination mechanisms, and the potential for performance
overhead contribute to the intricate nature of agent-based systems. The challenges extend
to aspects like debugging, troubleshooting, and adapting to dynamic environments,
making the implementation and management of agent-central architectures resource-intensive
and demanding careful design considerations [
A summary of the preceding points is depicted in Figure 3, where both the limitations and benefits of employing agents in software architectures can be observed. It is clear that the combination of agents and microservices ofers certain advantages that address the limitations found in the combination of agents and SOA.
To illustrate, if we encounter issues like scalability and resource usage in the context of agents combined with SOA, we can address them by taking advantage of the efective resource management and optimization features found in agents combined with microservices. Likewise, challenges related to distributed processing and interoperability, as discussed in the context of agents and SOA, can be resolved by utilizing the flexibility, autonomy, and adaptability ofered by agents combined with microservices. Moreover, the problem of limited reusability often associated with SOA combined with agents can be overcome by integrating and collaborating with agents combined with microservices, which enhances the agent’s capacity for improvement. Additional benefits are intrinsic to both types of architectures, as is evident from the earlier discussion, where microservices can be viewed as an evolution of SOA.
Finally, common limitations emerge when deploying agents in both architectures, implying persistent challenges. In the following section, we introduce MLSysOps, a project designed for automatic and adaptive resource management and application deployment. This initiative is poised to tackle some of the challenges associated with these limitations.
The MLSysOps project represents a crucial initiative aimed at navigating the complexities surrounding resource management and application deployment within DEC continuum systems, leveraging the power of ML techniques. It is conceived as a Platform-as-a-Service (PaaS) solution,dedicated to streamlining the management, and deployment of distributed IoT applications. These applications, comprising interconnected components with diverse resource demands and considerations for quality of service, find a unified platform in MLSysOps. The advantages explained in the previous section seamlessly extend to the MLSysOps project. In particular, the utilization of this agent framework to oversee microservice-oriented IoT applications within the DEC Continuum ofers a range of functionalities, encompassing autonomy, intelligence, lfexibility, and integration.
QoS
MLaaS M1 M2 M3 .....
NODE
Agent
Telemetries MAPE
ML Models
As illustrated in Figure 4, agents, equipped with their full range of functionalities, will be capable of internally and externally employing ML models on each node to oversee the execution of the application. Additionally, a ML-as-a-service repository will be available, allowing agents to choose from a variety of models to govern the entire system based on its current state. This approach efectively addresses and overcomes the dynamic challenges inherent in the ecosystem.
Furthermore, addressing the limitations within the MLSysOps project involves harnessing the power of ML techniques. The application of ML is instrumental in enhancing security by incorporating well-established anomaly detection models. This robust security measure complements the amalgamation of agents and microservices. Similarly, performance optimization and reduction of overhead are achieved through continuous monitoring system state and eficient manage of the resources facilitated by ML. This dynamic orchestration of services serves to preemptively mitigate potential performance constraints, ensuring a seamless operational experience.
While it is acknowledged that the implementation of such agents may demand a certain level of complexity and specialized skills, it is imperative to recognize that this limitation extends beyond system behavior. Emphasizing the evolutionary leap represented by this promissory advancement can efectively dispel concerns about complexity, encouraging broader adoption of this architecture.
In the era of expanding IoT ecosystems, where the inclusion of devices and services is a daily occurrence, efectively dealing with this vast, dynamic, and diverse environment poses a significant challenge. In this direction, the exploitation of software agents holds great promise to facilitate the automation of this complex management landscape. Our analysis has delved into the utilization of agents within SOA and microservice architectures, shedding light on their advantages and limitations.
Upon comparison, it becomes evident that incorporating agents into a microservices scenario aligns seamlessly with the distributed nature of the IoT. This integration brings forth essential functionalities such as autonomy in resource management, flexibility, adaptability, integration, and collaboration, which are instrumental in realizing the vision of autonomous computing. Nevertheless, certain limitations persist, presenting opportunities for refinement.
The MLSysOps framework, as presented, signifies a significant stride in addressing these challenges. The amalgamation of agents and microservices, coupled with the incorporation of ML techniques, promises to embed intelligence into the automation of the dynamic and heterogeneous DEC continuum. This integrated approach not only harnesses the inherent capabilities of agents and microservices but also addresses and surpasses the previously identified limitations.
The future work involves the practical implementation of the MLSysOps framework and rigorous validation through numerous tests using real-world distributed IoT applications. This real-world validation will not only attest to the feasibility of the proposed framework but also contribute valuable insights for further enhancement. As we navigate the future of IoT ecosystems, the synergistic integration of agents, micro-services, and ML techniques stands as a promising avenue for overcoming the intricacies of managing this ever-expanding landscape.
The research leading to this work was carried out under the “MLSysOps" Project (Grant Agreement 101092912) funded by the European Community’s Horizon Europe Programme. For the purpose of open access, the authors have applied a CC BY public copyright license to any Author Accepted Manuscript version arising from this submission. computing for autonomic computing: A research agenda, IEEE Internet Computing 13 (2009) 82–87. [12] P. Vrba, V. Mařík, P. Siano, P. Leitão, G. Zhabelova, V. Vyatkin, T. Strasser, A review of agent and service-oriented concepts applied to intelligent energy systems, IEEE transactions on industrial informatics 10 (2014) 1890–1903. [13] F. Siqueira, J. G. Davis, Service computing for industry 4.0: State of the art, challenges, and research opportunities, ACM Computing Surveys (CSUR) 54 (2021) 1–38. [14] Y. Kalyani, R. Collier, Hypermedia multi-agents, semantic web, and microservices to enhance smart agriculture digital twin, in: 2023 IEEE International Conference on Pervasive Computing and Communications Workshops and other Afiliated Events (PerCom Workshops), IEEE, 2023, pp. 170–171. [15] T. Daradkeh, A. Agarwal, Modeling and optimizing micro-service based cloud elastic management system, Simulation Modelling Practice and Theory 123 (2023) 102713. [16] R. W. Collier, E. O’Neill, D. Lillis, G. O’Hare, Mams: Multi-agent microservices, in: Companion proceedings of the 2019 world wide web conference, 2019, pp. 655–662.