[Paper Review] Enabling Community Health Care with Microservices

Richard Hill, Dharmendra Shadija+1 more|arXiv (Cornell University)|Sep 20, 2017

Software System Performance and Reliability17 references3 citations

TL;DR

This paper proposes a microservices-based IoT architecture for community health care that uses edge processing and workload management to reduce data transmission by 57% and compute time by 59%, enhancing scalability, resilience, and privacy in decentralized care systems. The approach leverages agent-oriented modeling and LoraWAN for efficient, secure data flow from home sensors to health clouds.

ABSTRACT

Microservice architectures (MA) are composed of loosely coupled, course-grained services that emphasise resilience and autonomy, enabling more scalable applications to be developed. Such architectures are more tolerant of changing demands from users and enterprises, in response to emerging technologies and their associated influences upon human interaction and behaviour. This article looks at microservices in the Internet of Things (IoT) through the lens of agency, and using an example in the community health care domain explores how a complex application scenario (both in terms of software and hardware interactions) might be modelled.

Motivation & Objective

• To address the growing complexity of community health care systems by enabling scalable, resilient, and privacy-aware service composition.
• To reduce network bandwidth and processing overhead in IoT-based health monitoring by filtering and processing sensor data at the edge.
• To explore the viability of microservices and agent-oriented modeling in supporting dynamic, multi-stakeholder care environments.
• To evaluate the performance gains of workload management in edge computing units for real-time health data processing.
• To identify opportunities for decentralized data processing and secure data sharing between health providers and local authorities.

Proposed method

• Designing a microservice-based architecture where each service encapsulates a specific care function, such as data filtration, analytics, and reporting.
• Implementing a two-stage data processing pipeline at the edge: initial filtering of sensor data followed by pattern recognition and trend analysis.
• Using Agent Unified Modeling Language (AUML) to model the system as a multiagent system, enabling proactive and coordinated behavior among stakeholders.
• Employing LoraWAN for low-power, long-range communication of non-critical data from home monitoring units to the health cloud.
• Applying workload management algorithms to dynamically distribute processing across edge components based on available resources and deadlines.
• Simulating the system using Omnet++ 5.1 on Ubuntu 16.04 with a 100MB sensor dataset from MIT AI Lab to evaluate performance under varying loads.

Experimental results

Research questions

• RQ1 How can microservices and edge processing reduce data transmission and improve efficiency in community health care IoT systems?
• RQ2 What is the impact of workload management on processing latency and resource utilization in edge-based health monitoring units?
• RQ3 How does a microservices architecture enhance system resilience and modularity in dynamic, multi-stakeholder health care environments?
• RQ4 To what extent can decentralized data processing reduce bandwidth consumption in constrained LoraWAN networks?
• RQ5 How can data privacy and access control be enforced across distributed microservices in community health care systems?

Key findings

• Data transmission was reduced by 57% across the simulated network through edge-based filtration and workload-aware processing.
• Compute time for processing sensor data was reduced by 59%, from 18.3 seconds to 7.3 seconds, due to optimized workload distribution.
• Network bandwidth consumption decreased by 51% as a result of filtering redundant sensor data before transmission.
• The system demonstrated improved resilience and adaptability under increasing sensor loads due to dynamic workload management.
• The edge architecture maintained performance stability even with high sensor input, showing reduced sensitivity to network fluctuations.
• The Home Monitoring unit’s computational capacity was underutilized, indicating potential for further optimization through advanced data processing.