Abstract: The Synthy Approach for End to End Web Services Composition: Planning with Decoupled Causal and Resource Reasoning Biplav Srivastava Web services offer a unique opportunity to simplify application integration by defining common, web-based, platform-neutral, standards for publishing service descriptions to a registry, finding and invoking them -- not necessarily by the same parties. Viewing software components as web services, the current solutions to web services composition based on business web services (using WSDL, BPEL, SOAP etc.) or semantic web services (using ontologies, goal-directed reasoning etc.) are both piecemeal and insufficient for building practical applications. Inspired by the work in AI planning on decoupling causal (planning) and resource reasoning (scheduling), we introduced the first integrated work in composing web services end to end from specification to deployment by synergistically combining the strengths of the current approaches. The solution is based on a novel two--staged composition approach that addresses the information modeling aspects of web services, provides support for contextual information while composing services, employs efficient decoupling of functional and non-functional requirements, and leads to improved scalability and failure handling. A prototype of the solution has been implemented in the Synthy service composition system and applied to a number of composition scenarios from the telecom domain. The application of planning to web services has also brought new plan and planner usability-driven research issues to the fore for AI.
Introduction: Bridging the Gap in Service Composition
In the modern business landscape—particularly in fast-moving sectors like telecommunications—there is a constant need to create new applications by stitching together existing web services (e.g., location tracking, billing, SMS). Historically, two silos existed: industry standards like WSDL and BPEL, which focus on execution but require manual workflow design, and academic AI planning, which focuses on logic but often ignores deployment realities and non-functional requirements (NFRs).
Synthy is an end-to-end framework that automates the composition of Web Services by bridging this gap. It takes a high-level goal and automatically produces a deployable, optimized execution plan.
Methods: Decoupling Causal and Resource Reasoning
The core innovation of Synthy is the decoupling of causal reasoning (planning) and resource reasoning (scheduling), inspired by the RealPlan AI approach. This is achieved through a two-staged composition process:
- Logical Composer (LC): This stage handles causal reasoning. It uses generative planning to create an abstract workflow based on functional requirements. Instead of searching through every individual service instance, it plans using Service Types defined in a domain ontology, which significantly reduces the search space.
- Physical Composer (PC): This stage handles resource reasoning. It takes the abstract plan and binds each service type to a specific Service Instance found in a registry. It uses multi-objective optimization to satisfy NFRs, such as selecting instances with the lowest latency or highest throughput.
Once the physical binding is complete, Synthy generates an executable BPEL (Business Process Execution Language) workflow, compatible with standard industry engines.
Telecom Domain Scenario
The framework was evaluated using a complex telecom use case: the Helpline Service. This scenario involves orchestrating components like Location Tracking, SMS Delivery, Call Setup, and Problem Classification.
The system demonstrated hierarchical composition, where a newly composed service (like a location-based agent selector) can be registered back into the system as a new "type" for future plans, enhancing reuse.
Key Insights and Results
- Abstraction is Key: By differentiating between service types (what it does) and instances (who does it), the planner avoids the "state-space explosion" common in large registries.
- AI as a Compiler: Synthy treats the planner as a high-level "compiler" for industry standards, turning abstract goals into low-level BPEL code.
- Scalability: The decoupled approach allows the system to handle large collections of web services more efficiently than traditional integrated planners.
Limitations and Future Work
While Synthy automates the pipeline from specification to deployment, challenges remain in planner usability and the quality of service metadata. Future work will focus on automatic parameter tuning for planners and improving real-time re-planning and re-binding to handle dynamic environmental changes, such as service outages.
Summary updated