The Internet of Things

Ubiquitous computing is the core of the Internet of Things, which means incorporating computing and connectivity in all the things around us. Interoperability of such heterogeneous devices needs well-defined standards. But standardization is difficult because of the varied requirements of different applications and devices. For such heterogeneous applications, the solution is to have a middleware platform, which will abstract the details of the things for applications. That is, it will hide the details of the smart things. It should act as a software bridge between the things and the applications. It needs to provide the required services to the application developers so that they can focus more on the requirements of applications rather than on interacting with the baseline hardware. To summarize, the middleware abstracts the hardware and provides an Application Programming Interface (API) for communication, data management, computation, security, and privacy.

The challenges, which are addressed by any IoT middleware, are as follows:

1. Interoperability and programming abstractions: for facilitating collaboration and information exchange between heterogeneous devices, different types of things can interact with each other easily with the help of middleware services. Interoperability is of three types: network, semantic, and syntactic. Network interoperability deals with heterogeneous interface protocols for communication between devices. It insulates the applications from the intricacies of different protocols. Syntactic interoperability ensures that applications are oblivious of different formats, structures, and encoding of data. Semantic interoperability deals with abstracting the meaning of data within a particular domain. It is loosely inspired by the semantic web.
2. Device discovery and management: this feature enables the devices to be aware of all other devices in the neighborhood and the services provided by them. In the Internet of Things, the infrastructure is mostly dynamic. The devices have to announce their presence and the services they provide. The solution needs to be scalable because the devices in an IoT network can increase. Most solutions in this domain are loosely inspired by semantic web technologies. The middleware provides APIs to list the IoT devices, their services, and capabilities. In addition, typically APIs are provided to discover devices based on their capabilities. Finally, any IoT middleware needs to perform load balancing, manage devices based on their levels of battery power, and report problems in devices to the users.
3. Scalability: a large number of devices are expected to communicate in an IoT setup. Moreover, IoT applications need to scale due to ever increasing requirements. This should be managed by the middleware by making required changes when the infrastructure scales.
4. Big data and analytics: IoT sensors typically collect a huge amount of data. It is necessary to analyze all of this data in great detail. As a result a lot of big data algorithms are used to analyze IoT data. Moreover, it is possible that due to the flimsy nature of the network some of the data collected might be incomplete. It is necessary to take this into account and extrapolate data by using sophisticated machine learning algorithms.
5. Security and privacy: IoT applications are mostly related to someone's personal life or an industry. Security and privacy issues need to be addressed in all such environments. The middleware should have built in mechanisms to address such issues, along with user authentication, and the implementation of access control.
6. Cloud services: the cloud is an important part of an IoT deployment. Most of the sensor data is analyzed and stored in a centralized cloud. It is necessary for IoT middleware to seamlessly run on different types of clouds and to enable users to leverage the cloud to get better insights from the data collected by the sensors.
7. Context detection: the data collected from the sensors needs to be used to extract the context by applying various types of algorithms. The context can subsequently be used for providing sophisticated services to users.

There are many middleware solutions available for the Internet of Things, which address one or more of the aforementioned issues. All of them support interoperability and abstraction, which is the foremost requirement of middleware. Some examples are Oracle's Fusion Middleware, OpenIoT, MiddleWhere, and Hydra. Middlewares can be classified as follows on the basis of their design:

1. Event based: here, all the components interact with each other through events. Each event has a type and some parameters. Events are generated by producers and received by the consumers. This can be viewed as a publish/subscribe architecture, where entities can subscribe for some event types and get notified for those events.
2. Service oriented: service oriented middlewares are based on Service Oriented Architectures (SOA), in which we have independent modules that provide services through accessible interfaces. A service oriented middleware views resources as service providers. It abstracts the underlying resources through a set of services that are used by applications. There is a service repository, where services are published by providers. The consumers can discover services from the repository and then bind with the provider to access the service. Service oriented middleware must have runtime support for advertising services by providers and support for discovering and using services by consumers. HYDRA is a service oriented middleware. It incorporates many software components, which are used in handling various tasks required for the development of intelligent applications. Hydra also provides semantic interoperability using semantic web technologies. It supports dynamic reconfiguration and self-management.
3. Database oriented: in this approach, the network of IoT devices is considered as a virtual relational database system. The database can then be queried by the applications using a query language. There are easy to use interfaces for extracting data from the database. This approach has issues with scaling because of its centralized model.
4. Semantic: semantic middleware focuses on the interoperation of different types of devices, which communicate using different formats of data. It incorporates devices with different data formats and ontologies and ties all of them together in a common framework. The framework is used for exchanging data between diverse types of devices. For a common semantic format, we need to have  adapters for communication between  devices because; for each device, we need adapters to map  standards to one abstract standard. In such a semantic middleware, a semantic layer is introduced, in which there is a mapping from each resource to a software layer for that resource. The software layers then communicate with each other using a mutually intelligible language (based on the semantic web). This technique allows multiple physical resources to communicate even though they do not implement or understand the same protocols.
5. Application specific: this type of middleware is used specifically for an application domain for which it is developed because the whole architecture of this middleware software is fine-tuned on the basis of requirements of the application. The application and middleware are tightly coupled. These are not general purpose solutions.