Towards a Practical Architecture for Internet of Things: An India-centric View

 

The current, widespread thinking on the Internet of Things (IoT) makes several (arguably misplaced) assumptions. IoT architectures are often a repackaging of existing ideas or a clean-slate, costly design. Some of these design assumptions include hundreds of tightly coupled devices; costly devices ($5 − $500) customized for an application; exclusive use of well-structured, always-on communication networks (e.g., IPv6); centralized data collection, analysis and control in the cloud; and a single vendor who owns the vertical application: platform, cloud services, data and ecosystem. Examples of these vertically integrated silos include smart power grids, supervisory control and data acquisition (SCADA) systems, and personal monitoring devices like FitBit.

However, an effective architecture for IoT, particularly for an emerging nation like India [1] with limited technology penetration at the national scale, should be based on: (1) tangible technology advances in the present, (2) practical application scenarios of social and entrepreneurial value, and (3) ubiquitous capabilities that make the realization of IoT affordable and sustainable. This is especially critical as India embarks on an ambitious program to upgrade 100 existing and new cities into Smart Cities. In this context, a rethink of the above assumptions suggests:

• Thousands of loosely connected devices in the immediate vicinity, and millions more further out;
• Ultra-cheap devices ($0.01 − $3.00) combined with existing generic, in-person devices like smartphones;
• A mix of ad hoc P2P, 2G/3G/4G, IPv6 and WiFi-based networks having intermittent connectivity;
• Data collection and personalized analytics that seamlessly span edge devices and the cloud, with control over data sharing and ownership while encouraging Open Data;
• A vendor and domain-neutral open ecosystem using internet/web standards, allowing reuse of devices and data.

Here, we propose ten design paradigms to achieve a sus­tainable and practical IoT architecture for India.

1) Human-centric rather than thing-centric design

Current IoT architectures are device or network oriented. How­ever, the key value proposition of IoT is from the interaction of 'Things' with humans and society; and the benefits gained for humans who are part of, affected by and influence the network. Technologies, services and decision making must create an IoT experience that deeply engages with people. This may be mundane, like providing optimal traffic routing in a smart city; or essential, like offering personalized health suggestions for patients. As a result, we favor devices, networks, data, and analytics that are in close proximity to humans and are widely prevalent, e.g., smart and feature phones, Bluetooth/WiFi/ 2G/3G /4G wireless interfaces, wearables/body area networks.

2) Span virtual and physical worlds

Much of the IoT conversation is about the physical infras­tructure and its optimization. Bringing in humans and social elements (with their virtual online avatars such as social networks and virtual agents) helps span the digital and physical worlds, and also integrate across humans and infrastructure. Capturing proximity and interactions between humans and 'Things' (H2H, H2M, M2M), both in the physical and virtual worlds, is necessary for actionable intelligence.

3) Big-little data

Analytics performed on information from diverse sources within the IoT architecture helps with data-driven decision making. There are two classes of such data: (1) transient sensor and personal data collected continuously from hu­mans/physical devices, i.e., 'Little' data; and (2) persistent knowledge-bases and archives that span domains and available in central repositories/clouds, i.e., 'Big' data. Meaningful analytics requires both 'Big' and 'Little' data to be combined, and often in real time.

4) Analytics from the edge to the cloud

Related to Big-Little data is performing distributed analytics and decision making. The current model of pushing all data to a central cloud for analytics will not scale, is inefficient, and raises privacy concerns. Given the enhanced capabilities of edge devices like smartphones coupled with intermittent network connections, decisions on whether a subset of the Big data and decision-analytic should be pushed to the phone, or the Little data and analytic aggregated in the cloud have to be automated. These are informed by the device capability, privacy needs, energy and network costs, and application QoS.

5) Bring the network to the sensor

As tens of thousands of cheap IoT devices proliferate, they will be constrained in energy and communication capabilities. Rather than rely on massive deployment of custom sensor networks and new standards, there is value in piggybacking on existing, widely adopted standards and reusing symbiotic infrastructure. For example using smartphones as P2P data mules for last mile connectivity to sensors, combined with highly functional gateways and clouds for coordination, suggests an asymmetric architecture.

6) How 'low' can you go?

Technology penetration has not been uniform across countries, regions, or, for that matter, industries. This disparity is a reflection of the differences in infrastructure, cost of access, telecom networks and services, and policies among different economies. Hence, the cost and technology behind the sensing, device, networking and analytic solutions for the IoT should be affordable and scale to billions of users. This requires reuse of commodity hardware and sensors, and existing infrastructure in novel ways rather than custom solutions with cutting-edge capabilities, or canned solutions developed for advanced economies. The cost-to-benefit trade-offs become critical.

7) Whose data is it anyway?

The intersection of devices, communication, data and humans within IoT offers interesting incentives and business models. A key success of the WWW is the ability for businesses to monetize users’ data (e.g., Google Ads using user’s web data pays for free search and email services). With IoT, devices are going to be even closer to humans and blend into our environment. Ensuring transparency in data ownership, sharing, and usage are important. Further, there is scope for data brokering that encourages open data sharing by users with business in return for clear rewards, be they monetary, peer recognition, or for the greater good.

8) When is 'good enough' enough?

IoT is naturally a diverse ecosystem with unreliability and uncertainties as: (1) cheap sensors mean questionable data quality, (2) humans are fickle to model, (3) physical systems are complex, (4) distributed 'Things' and intermittent com­munication are a given, and (5) data privacy puts bounds on its availability. As a result, analytics and decision making have to be probabilistic; and the system and application has to be conscious of what is 'good enough' and not fail in the absence of perfect behavior.

9) Context determines the action

Given the uncertainties of the system and humans being central entities, much of the decision making within the IoT infrastructure and applications has to be contextual. Context binds people and things to a common scope, and hence, will ease mining of relevant information. There has to be semantic knowledge that captures system and social behavior, some specified while others are learned using models. Intelligent agents will often act on the behalf of humans. They may be aware of personal preferences (e.g., Apple’s Siri, Microsoft’s Cortana, and Google Now), and these will interact with digital agents of service providers, utilities and vendors. Semantic context will have to complement web standards for structural syntax to allow such M2M interaction to be effective.

10) Business canvas

If the IoT is to yield successful business models, we first need to recognize that IoT is not a new product or market. What IoT brings is an additional set of technologies, lower power, more computation and storage, cheaper devices, better wireless connectivity, much more granular control and observation capabilities. What it enables is scaling in both directions – up and down, and the ability to look at ourselves and the world in an unprecedented degree of detail.

IoT business models fall into two broad categories: (1) hori­zontals, concerned with enabling components and technology; and (2) verticals, which integrate these technologies to supply an end user with a value proposition. The first set of horizontal business models is the development of specific sensors and actuators that enable the generation of new or more cost effective observations. The second model is the deployment horizontal, a business model that addresses the needs of building out to scale of the data gathering, data storage and data curation and data brokering needs of IoT based systems. The third horizontal business model addresses the needs for a portfolio of analytical techniques to convert the data gathered into actionable information. While the first two have been the focus of IoT’s precursor technologies, IoT’s scale is driving active development across the board.

Verticals will pull solutions and services across these horizon­tals to deliver final end customer value. The emphasis here will be on the necessary domain and system integration expertise and the ability to build the necessary collaborations across customers and suppliers. Likely verticals in developing countries will be based on condition monitoring needs for infrastructure, especially in energy and water. While healthcare also seems to be a popular topic, the most likely focus will be on device- based models.

Conclusion

In this article, our attempt has been to identify the core problems in the existing IoT ecosystem, and propose design paradigms that will make the realization of IoT afford­able and sustainable. For developing nations such as India, with their unique set of infrastructure challenges, the solutions needed are also different. Through such innovative solutions informed by our design proposals, developing nations can potentially leapfrog into more advanced IoT outcomes, just as the 2G and 3G telecom revolution in India skipped past pervasive landline telephones.

 

References

[1] Internet of Things (IoT): how to tap the $12 billion market in India. http://goo.gl/m6ui7X.

 

---

 

Prasant Misra is senior member, technical staff at the Robert Bosch Centre for Cyber Physical Systems in the Indian Institute of Science, Bangalore. He performs research in low-power sensing, wireless communication, and energy-efficient computing with a focus on system design and implementation within the general framework of Cyber Physical Systems and Internet of Things. He received his Ph.D. in Computer Science and Engineering from the University of New South Wales, Sydney. Currently, he serves as an associate technical editor of IEEE Communication Magazine, and is a member of IEEE and ACM.

 

Yogesh Simmhan is an Assistant Professor at the Supercomputer Education and Research Centre at the Indian Institute of Science, Bangalore. Previously, he was a Research Assistant Professor of Electrical Engineering and Associate Director of the Center for Energy Informatics at the University of Southern California, Los Angeles. His research explores abstractions, algorithms and applications on distributed data and computing systems, to advance fundamental knowledge and to offer a practitioner's insight, on building scalable systems for Big Data applications. He has a Ph.D. in Computer Science from Indiana University and is a Senior Member of IEEE and ACM.

 

Jay Warrior has over 20 years of experience creating new, high technology based, business opportunities for Agilent Technologies, Hewlett-Packard, Fisher-Rosemount, and Honeywell in the US and in Asia. His expertise covers the whole innovation cycle from market opportunity identification through strategy, business model and technology development, to building and managing execution teams for the new business. A senior member of IEEE, he has a long association with the development of IEEE standards for industrial automation and sensors.

 

Comments