Traditional software engineering relies heavily on static integration: components are bound together at compile-time or hardcoded during deployment. However, as applications have scaled across the Internet, the Internet of Things (IoT), and cloud-native environments, static architectures struggle under the weight of changing contexts, heterogeneous devices, and unpredictable runtime conditions.
To solve this, researchers developed the paradigm of dynamic service composition. Instead of treating software as a rigid machine, this approach treats applications as fluid assemblies of independent, atomic services that are discovered, configured, and chained together on-the-fly.
As noted in the foundational framing of "Internetware" by Gang Huang et al. in their Editor's note, software operating in open, chaotic environments must become autonomous and situational. Instead of executing a fixed script, systems must dynamically discover and integrate external services at runtime to fulfill high-level functional goals. This approach scales from enterprise stacks utilizing vertical service composition via an Enterprise Service Bus (Jun Shen et al. in Guest editorial notes for selected papers from SOCA 2010) to specialized domains like digital photogrammetry, where complex workflows are assembled out of individual, distributed web services (Kyoungjin Park et al. in A Design of Web Service for Digital Photogrammetry Workstation using Service Oriented Architecture).
The Mechanics of Indirection and Chaining
To understand why dynamic composition works, we must look at the underlying routing mechanism. How do independent services talk to each other without knowing their counterparts in advance? The answer lies in indirection and identifier stacking.
In the Internet Indirection Infrastructure (i3) proposed by Ion Stoica et al. in Internet Indirection Infrastructure, senders are completely decoupled from receivers through a rendezvous abstraction. Packets carry a stack of identifiers rather than a single destination IP address. As a packet traverses the network, each identifier can trigger a specific operation—such as data transcoding or localized routing—before forwarding the packet to the next stage. This chaining mechanism allows complex, multi-stage workflows to emerge naturally from independent network nodes.
Building on this network-centric view, Markus Keidl et al. introduced ServiceGlobe: Distributing E-Services Across the Internet, a platform where composite services are defined abstractly and then dynamically deployed onto arbitrary servers across the internet. By prioritizing distributed execution and intelligent service selection over fixed server infrastructure, the system can instantiate and execute workflows wherever resources are most available.
Context-Awareness and Semantic Planning
Dynamic composition is not merely about connecting pipes; it requires the system to understand what to connect and when. In pervasive and ubiquitous computing, this decision-making is driven by semantic metadata and real-time environmental triggers.
Consider a smart environment where a user walks into a room. The system must instantly orchestrate available hardware and software resources to accommodate them. Sushil Puradkar et al. developed the SEM-Ether framework in SEM-Ether: Semantic Web Based Pervasive Computing Framework - Integrating Web, Devices and People, which treats physical devices as atomic services described by semantic ontologies. A centralized service planner dynamically chains these services into a customized workflow based on the user's web-hosted profile and real-time triggers. Similarly, Yugyung Lee et al. demonstrated in Web-Based Semantic Pervasive Computing Services how automated service flows can be constructed using a rule-based composer that evaluates spatial data and user status to link heterogeneous web services appropriately.
This high degree of dynamism changes how we manage software lifecycles. As Heinz-Gerd Hegering et al. argue in Management Challenges of Context-Aware Services in Ubiquitous Environments, context-aware systems require a "micro-lifecycle" where service instances are created, configured, and terminated in an entirely ad-hoc manner. These transient services exist only as long as specific contextual conditions are met, making real-time discovery and composition essential. This is particularly critical in high-stakes, rapidly evolving scenarios, such as the tactical environments explored by Paul Thompson in Dynamic Integration of Distributed Semantic Services: Infrastructure for Process Queries and Question Answering, where static integration fails and information services must be dynamically integrated to support real-time decision-making.
Virtualization and En-Route Computation
The principles of dynamic composition have heavily influenced modern networking through Software-Defined Networking (SDN) and Network Function Virtualization (NFV). Here, the "services" being composed are virtualized network functions (VNFs) like firewalls, routers, or data processors.
Instead of routing data to a centralized data center for processing, modern networks perform "en-route" computation. In the CONDENSE architecture proposed by Dejan Vukobratovic et al. in CONDENSE: A Reconfigurable Knowledge Acquisition Architecture for Future 5G IoT, SDN and NFV are used to dynamically orchestrate and chain atomic computational modules across distributed 5G IoT nodes. By calculating functions (such as neural network layers or data averages) as data travels through the network, the system drastically reduces bandwidth pressure.
To make these networks resilient, Muhammad Imran et al. highlighted in IEEE Access Special Section Editorial: Survivability Strategies for Emerging Wireless Networks that service function chaining can be enhanced by incorporating environmental and network context to dynamically order and place VNFs. This allows for adaptive routing service customization, which Chao Bu et al. demonstrated in Enabling Adaptive Routing Service Customization via the integration of SDN and NFV. By integrating centralized network control with virtualized functions, the network can adaptively assemble basic and special-purpose functions on-the-fly to meet the granular requirements of diverse applications.
AI Agents and Software-Defined Systems
Today, the concept of dynamic service composition has expanded beyond web APIs and network packets into the realms of artificial intelligence and cyber-physical systems. The vision of marrying AI with service-oriented computing, as framed by Wenjun Wu 0001 et al. in Message from General Chairs of IEEE AISA 2020, focuses on automating the assembly of heterogeneous, cross-domain microservices to handle the scale and variability of cloud-native applications.
This is vividly apparent in modern generative AI architectures. In the MultiFluxAI platform introduced by Raveendra Kumar Medicherla et al. in Workshop Report on Generative AI-based Software Engineering (GenAI4SE 2nd Edition), an intelligent orchestration framework manages multiple AI agents. These agents dynamically select and invoke appropriate AI services and knowledge bases based on user queries and real-time context, moving past rigid, static Retrieval-Augmented Generation (RAG) pipelines. Taking this a step further, Amin Beheshti proposed Natural Language-Oriented Programming (NLOP) in Natural Language-Oriented Programming (NLOP): Towards Democratizing Software Creation, where an intelligent gateway leverages contextual understanding of microservices to dynamically manage traffic, rate limiting, and protocol translation, translating natural language intent directly into an operational microservice architecture.
Even hardware-heavy industries like automotive engineering are adopting this approach. The AUTOFRAME framework, developed by Sven Kirchner et al. and detailed in AUTOFRAME - A Software-driven Integration Framework for Automotive Systems, transitions vehicles to "Software Defined Vehicle" (SDV) architectures. A centralized connection handler automatically manages data flow between isolated software containers (like lane detection or motion planning) based on a global configuration file. This allows vehicle features to be dynamically added, updated, or re-composed at runtime without requiring a full system reboot or manual hardware reconfiguration.
Why Dynamic Composition Works
Dynamic service composition works because it shifts the responsibility of integration from the developer at design-time to an intelligent orchestration layer at runtime. By decoupling components through indirection, describing them with semantic metadata, and leveraging virtualization, systems can continuously reconfigure themselves to match the exact situational needs of the user, network, or environment.
Key Foundational Work
ServiceGlobe: Distributing E-Services Across the Internet (2002) — contributed 3 concept elements to this direction.
The Ninja architecture for robust Internet-scale systems and services (2001) — contributed 2 concept elements to this direction.
Internet Indirection Infrastructure (2002) — contributed 1 concept elements to this direction.