Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLM-based policy generation for intent-based management of applications

Kristina Dzeparoska, Jieyu Lin, Ali Tizghadam, Alberto Leon-Garcia

TL;DR

This work introduces Emergence, an intent-based management system that uses few-shot learning with large language models to progressively decompose user intents into a policy-based abstraction. The policies are structured as $\vec{P}=(D, E, A, \vec{C})$ with $\vec{C}=(\vec{R}, \vec{T}, \vec{S})$, and mapped to APIs through policy-to-API and MEF PDO translations, forming a closed $MAPE-K$ loop for deployment. The approach is validated on a VNF service-chain use-case (DPI, load balancer, web servers) in a OpenStack/SAVI environment, demonstrating both intent fulfillment and assurance, including remediation when drift is detected. Key findings indicate that generic LLMs can generalize to unseen intents with few-shot prompts, enabling automatic decomposition into executable policies and revealing the potential to extend policy execution through dynamic code and FSM generation. The work highlights practical implications for scalable, autonomous application management and points to future directions in validation, digital twins, and broader language-to-policy conversions.

Abstract

Automated management requires decomposing high-level user requests, such as intents, to an abstraction that the system can understand and execute. This is challenging because even a simple intent requires performing a number of ordered steps. And the task of identifying and adapting these steps (as conditions change) requires a decomposition approach that cannot be exactly pre-defined beforehand. To tackle these challenges and support automated intent decomposition and execution, we explore the few-shot capability of Large Language Models (LLMs). We propose a pipeline that progressively decomposes intents by generating the required actions using a policy-based abstraction. This allows us to automate the policy execution by creating a closed control loop for the intent deployment. To do so, we generate and map the policies to APIs and form application management loops that perform the necessary monitoring, analysis, planning and execution. We evaluate our proposal with a use-case to fulfill and assure an application service chain of virtual network functions. Using our approach, we can generalize and generate the necessary steps to realize intents, thereby enabling intent automation for application management.

LLM-based policy generation for intent-based management of applications

TL;DR

This work introduces Emergence, an intent-based management system that uses few-shot learning with large language models to progressively decompose user intents into a policy-based abstraction. The policies are structured as with , and mapped to APIs through policy-to-API and MEF PDO translations, forming a closed loop for deployment. The approach is validated on a VNF service-chain use-case (DPI, load balancer, web servers) in a OpenStack/SAVI environment, demonstrating both intent fulfillment and assurance, including remediation when drift is detected. Key findings indicate that generic LLMs can generalize to unseen intents with few-shot prompts, enabling automatic decomposition into executable policies and revealing the potential to extend policy execution through dynamic code and FSM generation. The work highlights practical implications for scalable, autonomous application management and points to future directions in validation, digital twins, and broader language-to-policy conversions.

Abstract

Automated management requires decomposing high-level user requests, such as intents, to an abstraction that the system can understand and execute. This is challenging because even a simple intent requires performing a number of ordered steps. And the task of identifying and adapting these steps (as conditions change) requires a decomposition approach that cannot be exactly pre-defined beforehand. To tackle these challenges and support automated intent decomposition and execution, we explore the few-shot capability of Large Language Models (LLMs). We propose a pipeline that progressively decomposes intents by generating the required actions using a policy-based abstraction. This allows us to automate the policy execution by creating a closed control loop for the intent deployment. To do so, we generate and map the policies to APIs and form application management loops that perform the necessary monitoring, analysis, planning and execution. We evaluate our proposal with a use-case to fulfill and assure an application service chain of virtual network functions. Using our approach, we can generalize and generate the necessary steps to realize intents, thereby enabling intent automation for application management.
Paper Structure (17 sections, 2 equations, 8 figures, 1 table)

This paper contains 17 sections, 2 equations, 8 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Methodology
  4. Policy model functional abstraction
  5. Intent Decomposition using few-shot learning
  6. Intent Classification Stage
  7. Progressive intent decomposition Stage
  8. Intent Validation Stage
  9. Evaluation and Use-case
  10. Intent Fulfillment
  11. Intent Assurance
  12. Evaluation Results and Analysis
  13. Discussion and Opportunities
  14. LLMs for additional conversions
  15. Lessons Learned
  16. ...and 2 more sections

Figures (8)

  • Figure 1: Pipeline overview: 1) classify intents to known intent types, 2) progressively decompose intents and generate policies: map each policy to an API, execute the API and return the result to the LLM, 3) validate the policies. The resulting policy tree represents the sequence of derived policies from the intent.
  • Figure 2: Example intent to intent type classification for the first stage.
  • Figure 3: Example policy actions, resources, and constraints.
  • Figure 4: Few-shot training example for progressive decomposition of an Intent and its corresponding Intent type.
  • Figure 5: Progressive policy generation and execution for intent fulfillment. Intent: "Deploy a service function chain with high availability in Domain1 consisting of: a medium vm for the dpi service, a medium vm for the load-balancer service, and 2 small vms for the web servers". Intent type: create resource, deploy service, availability.
  • ...and 3 more figures