A Middleware Infrastructure for Utility-based Provisioning of IoT Cloud Systems

Summary

The authors devised a middleware that 1) Enables development of generic, multi-level provisioning workflows and 2) Supports automated and scalable execution of such workflows in IoT based Cloud systems. The main features of the middleware include:

1) A generic, light-weight resource abstraction mechanism, based on software-defined gateways, which enables application specific customization of Edge devices

2) Support for automated provisioning of Edge resources and application components in a logically centralized manner, via dynamically managed APIs;  

3) Flexible provisioning models that enable self-service, on-demand consumption of the Edge resources.

The authors demonstrated the way middleware supports on-demand, self-service resource consumption by providing flexible provisioning models. The middleware also supports uniform, logically centralized provisioning of Edge devices, application artifacts and their configuration models. They introduced provisioning support for software-defined gateways. These SDGs enable application-specific customization of Edge devices through well-defined APIs. Since these SDGs are virtualized entities they also leverage the advantages of virtualization. The evaluation shows that the middleware enables scalable execution of provisioning workflows across relatively large IoT cloud resource pool. while the resource consumption overhead is suitable for edge devices.

Strengths

The implementation details are extremely clear. The authors have done a great job of explaining each component of the framework in great detail, at the same time they provided valid reasons for using various entities in the paper.  

This framework also addresses scalability concerns by enabling microservices to be installed using different VMs. This enables controller to elastically scale up and and support multiple workflows.   

Although the idea of provisioning and management of devices is old, but on earlier frameworks it is based on tightly-coupled provisioning models such as static templates. On the other hand, in this paper authors have introduced automatic provisioning of IoT Cloud resources by implementing multi-level provisioning using provisioning workflows, which enable providing comprehensive provisioning support on multiple levels such as infrastructure-, platform- and application-level which makes it dynamic.

Older approaches supported specific data-centric tasks such as integrating various data formats, providing data-linking solutions and supporting communication protocols. Contrary, to them, this approach provides support for multi-level provisioning and consuming both IoT and Cloud resources as general-purpose utilities.

This framework takes into account the dependence of application business logic on underlying devices which generic cloud computing frameworks ignore. There is strong dependence of IoT Cloud applications on specific properties of the underlying Edge devices (e.g., available sensors) and novel resource features, which also needs to be considered during application provisioning.

Weaknesses

The middleware resource allocation framework currently only considers static device meta-information.

There was no mention of fault-tolerance approaches in case of device failures.

Security concerns in case when a user tries to breach in by executing malicious code.  

Points of Discussion:

How to gracefully handle device crashes ?

How to optimize placement of applications on Edge devices?  

How to handle device config changes during workflow provisioning ?

How to provide more dynamic and finer-grained resource monitoring ?