IoT: Respect the Present and Build for the Future

[This post was authored by Amit Jain (Director, Product Management). Special thanks to Satyam Vaghani (VP/GM for IoT and AI) for reviewing and contributing to this article]

Every time there is a fundamental change in the computing paradigm, as in IoT, there is an opportunity to rethink how can we respect the present and at the same time build for the future. This is what happened as we moved from Desktop to Smart Phones to Smart Watches.

The apps remained more or less the same — there was still an email client and a browser —
however the interaction and SLAs became fundamentally different, which led to the emergence of a new OS.

Extending beyond this led to the emergence of a new platform, which enabled the virtualization and consolidation of multiple apps foregoing separate silos. Think about cameras, watches, calculator etc. And with a single click, you could freely move the applications and data in and out of the cloud. The underlying substrate became invisible, which is the most important tenet of a simple yet powerful platform!

Lack of a platform is what’s preventing to unlock the full benefits of the IoT in major verticals. IoT provides tremendous benefits in the Oil and Gas Value Chain — upstream (improve drilling productivity and optimize simulation runs), midstream (provide operational efficiency and predictive maintenance) and downstream (optimize delivery and enhance end-user experience). However, these represent use-case specific IoT App silos with limited data mobility or re-use. Perhaps even proprietary apps with data lock-in. So, developing the next app becomes an arduous task, limiting cross-app insights while destroying the ROI.

And this issue needs to be addressed while respecting the WAN connection, which is over a flaky satellite link with high latency and low bandwidth. So, the Cloud-only “Alexa-model”, whereby the large volumes of sensor data go all the way to the cloud for both long-term and real-time processing, just won’t work. Any insights generated will be stale as the field workers need to act upon the condition of oil wells in real-time. So, compute needs to be moved closer to the sensors at the Edge.

However, Edge cloud, with local appliance connected to sensors, is incredibly difficult to operationalize. This is because of the diversity of sensors, which communicate via esoteric protocols — Modbus, CAN bus, PROFINET etc — and require different physical interfaces. The scale of deployment — involving 100s of depots to 1000s of oil rigs — makes it even more challenging.

Moreover, the next-gen Cloud Native apps require new constructs and various ML (Machine Learning) frameworks. Apps need to be able to run on myriad devices with various types of CPU — i7, Atom, Xeon-D etc. — and with various types of GPU — ASICs, FPGAs, add-on card from various vendors. Last, but not the least, there is the inherent complexity of human element as IT, OT, Developers and Data Scientists all need to come together to design, develop, deploy and operate the IoT app.

All in all, there is pure infrastructure madness, leaving little time or resources to be focused on business logic and outcome.

And so, the Chaos Reigns. Now, let’s Rein in the Chaos!

The new thinking warrants defining the attributes of the desired Platform. Before doing that, let’s first look at the demographics and their needs. There are two very important stakeholders in this food chain:

First are the Developers (#OnMyTerms), who are very opinionated. They like the convenience of SaaS Management Plane and the flexibility to bring their preferred Cloud as well as the ML models developed in any domain. Moreover, access to rich data/runtime services is must-have to execute Machine Inferencing at the Edge. They want the extensibility of programming frameworks, rich/open APIs, integration with CI/CD pipeline, first-class IDE and easy debuggability.

Second are the Operators (#GetMyWeekendBack), who just want to get things done quickly. They want to consolidate and reduce the infrastructure sprawl so as to eliminate the app silos. They want management simplicity for Planet Scale Operations, Zero-Touch onboarding with the ability to handle disconnected operations. Above all, the system should have built-in infrastructure and data security with Multi-tenancy.

What suits both is the ability to run Containers and leverage Cloud Native frameworks. Developers no longer need to design monolithic apps, but can utilize microservices framework to create building blocks which can then be leveraged across other apps. Easy Portability of Apps becomes a huge convenience. Operators have less layers to manage, and can patch/upgrade easily.

The good news is that at Nutanix, we incorporated the afore-mentioned considerations to design the new IoT Platform for you!

Introducing Xi IoT Platform

The Xi Edge OS is built on CentOS/Kubernetes, exposing Containers interface which allows you to consolidate the traditional IoT apps as well as enable the creation of new-generation, data science based apps built with Cloud-Native technologies. The Edge stack offers micro-PaaS (Platform-as-a-Service), a scaled down version of rich data and runtime services, which were otherwise available only in the Cloud, so that you can now implement Machine Inferencing at the Edge.

The platform provides Secure access to IoT data sources with Data Pipelines all the way from the Edge to any Cloud. The platform offers a choice of clouds, and allow seamless data mobility between edge and cloud, so that you can send metadata and build the ML models in the cloud. And not just Public clouds but Private Clouds are possible as well, based on policy or regulation in certain geos.

Just as Nutanix has converged many difficult yet significant problems in the datacenter, we have converged significant technology areas into a simple and delightful data processing platform for Edge and IoT apps.

How to Get Started

You can begin by Containerizing all your apps, legacy as well as the next-gen data science apps. The next step would be to inject the Yaml specification onto the Edge, which will pull the images from any Container Registry. It’s that easy, there are no extra manual steps of importing binaries or going through an arduous implementation guide. The Apps can be monitored through the convenience of SaaS management plane. And, this is CI/CD integrated. Developers will dig it!

The platform offers first-class Container-as-a-Service, which hides away the underlying K8S complexity and orchestration. Operators can easily manage planet-scale operations and enable zero-touch onboarding.

Note that you have the utmost deployment flexibility because Xi Edge OS can run as VM on your existing virtualized appliance (as may be the case for brownfield deployments whereby there could be some legacy VMs) or as bare metal on specialized HW, ready for any environment (greenfield, thus optimizing the resource usage on the edge device).

As a next step, you can deconstruct the containers to leverage more functionality that is built-in natively within the platform.

#ProgrammableEdge: Leverage Functions

And what’s really differentiating is that developers can now interact with the Edge platform through Functions (written in Python | Node.js | Golang), which are just focusing on the business outcome without worrying about infrastructure logic and transform the incoming sensor data. The developers don’t need to be concerned anymore as to where the sensor data is coming from or to re-compile depending on which Edge the app is getting pushed out to. The Platform middleware and runtime will get the job done!

#ExtensibleEdge: Leverage Data Pipeline with Streaming Analytics

Think of it as LEGO building blocks, whereby you can apply multiple, cascaded ML models to analyze the incoming data, transform it and send it across the Cloud (or store locally within the Edge), thus creating a Data Pipeline. You can #BYOModel or we can build one for you. The platform will have pre-built runtime environments with ML frameworks. At the same time, you can easily import any custom runtime, requiring specific version or libraries as per your needs. We optimize the Inferencing through model pruning, quantizing etc. in addition to leveraging TensorRT.

Examples

Gas leak detection ML models can be simply applied for pipeline monitoring and seamless midstream operations.

Similarly, through easy injection of license plate recognition ML model, you can give personalized retail experience and automated credit card processing for your end customers at the gas locations.

The possibilities — from upstream to downstream — are endless once you have established an agile platform for your business needs!

Takeaway

And this is the power of the platform approach, so that while respecting the Present, you can build for the Future. The blog talks about Oil and Gas Value chain challenges and benefits, however the concepts are generic so those can be applied to any other vertical as well. The platform makes IoT infrastructure invisible and enables rapid IoT app development and deployment, so your developers and operators can have the Freedom to Build, Run, Cloud, Invent and Play!

Please visit www.Nutanix.com/IoT for the latest information. You can also reach out to your local Nutanix sales team for detailed info.

Fun Fact: The United States is now the largest global crude oil producer. U.S. crude oil production, particularly from light sweet crude oil grades, has rapidly increased since 2011 and will continue to do so for 2019. The oil price decline in mid-2014 resulted in U.S. producers reducing their costs and temporarily scaling back crude oil production. However, after crude oil prices increased in early 2016, investment and production began increasing later that year. By comparison, Russia and Saudi Arabia have maintained relatively steady crude oil production growth in recent years.