MUSA: A Platform for Data-Intensive Services in Edge-Cloud Continuum

Marco Anisetti, Claudio A. Ardagna, Massimo Banzi, Filippo Berto, Ruslan Bondaruc, Ernesto Damiani, Alessandro Pedretti, Arianna Pisati, Antonio ReticoAina 2024 · DOI

Introduction

The traditional paradigm of centralized cloud computing is increasingly inadequate for modern, data-intensive applications requiring low latency and high throughput. As we transition toward an Edge-Cloud Continuum, computation is distributed across a heterogeneous landscape of on-premises devices, 5G edge nodes, and public clouds. However, managing this continuum introduces significant complexities regarding resource placement, non-functional requirements (security, privacy, QoS), and the handling of massive datasets.

The MUSA (Multilayered Urban Sustainability Action) platform is designed to address these challenges, providing a seamless environment for orchestrating complex workflows across the frontier of edge and cloud infrastructure.

Architectural Overview

MUSA acts as a central management layer that decouples application logic from the underlying physical infrastructure. By doing so, it allows developers to specify what an application does and what it needs (in terms of Quality of Service or privacy) without needing to manually configure specific hardware nodes.

Figure 1: High-level interaction between the MUSA platform and the Edge-Cloud Continuum.

The platform integrates several key components:

  • MUSA Platform & Data Hub Framework: A specialized service layer that manages data lakes and ensures privacy through techniques like homomorphic encryption and differential privacy.
  • 5G MEC Integration: MUSA explicitly leverages 5G Mobile Edge Computing (MEC) nodes. These nodes serve as a critical "middle ground," offering the reliability of managed infrastructure with the low latency of local edge computing.
  • Continuous Verification: Unlike static deployment models, MUSA monitors QoS and security properties in real-time, allowing for dynamic reconfiguration if performance targets are not met.

Deployment and Orchestration

At the heart of MUSA is the Deployment Engine, which functions as the system's "brain." It processes an Annotated Service Composition Template—a graph representing application requirements—and maps it against an Annotated Continuum Facilities Graph, which represents available hardware capabilities.

Figure 4: The MUSA Deployment Engine and its interaction with distributed deployment agents across the continuum.

The execution of these mappings is handled by Deployment Agents. These lightweight software components reside on each facility within the continuum (Cloud, 5G-Edge, and On-premises). They receive "deployment recipes" from the engine to configure and launch services locally, ensuring that the specific needs of each workflow step—such as high-speed processing or data locality—are met.

Case Study: Molecular Docking in Life Sciences

To validate the architecture, the authors implemented a real-world scientific workload: a Molecular Docking workflow. This process is essential for drug discovery but is computationally expensive and data-intensive.

The proof of concept utilized a dataset of 485 proteins from the Protein Data Bank (PDB). The MUSA-managed workflow automated several stages:

  1. Format Verification: Ensuring molecular structures are in the correct 3D format.
  2. Format Conversion: Dynamically converting 2D structures where necessary.
  3. Parallel Docking: Executing hundreds of simulations in parallel across the continuum.

By leveraging AWS Step Functions and containerized microservices, MUSA transformed a manual, infrastructure-heavy research task into a robust, automated "one-click" virtual screening tool.

Conclusion and Future Work

MUSA demonstrates that 5G is more than a connectivity standard; it is a fundamental architectural layer that enables "computation locality." By bridging the gap between local privacy and cloud scalability, MUSA provides a path forward for data-intensive science.

Future development will focus on refining the algorithms used for "recipe generation" and exploring optimization strategies for the deployment engine. Additionally, addressing the challenges of multi-provider environments remains a key priority to ensure the platform's viability in fragmented global markets.

Summary updated

BibTeX
@inproceedings{anisetti2024musa,
  author = {Marco Anisetti and Claudio A. Ardagna and Massimo Banzi and Filippo Berto and Ruslan Bondaruc and Ernesto Damiani and Alessandro Pedretti and Arianna Pisati and Antonio Retico},
  title = {{MUSA: A Platform for Data-Intensive Services in Edge-Cloud Continuum}},
  booktitle = {Aina},
  year = {2024},
  doi = {10.1007/978-3-031-57931-8\_32},
}
Figure