Where jellyfishes work better than whales

Edge Distributed Computing is revolutionising supercomputing, offering agility, cost-effectiveness, and real-time decision-making.

In a world besotted with big-bellied supercomputers and specially-encased Quantum wonders, it is interesting to consider that there are alternatives to handle huge computing workloads without worrying about space, fragility and tonnes of metal. It is an age where IoT devices, sensors and equipment-attached computing brains can be deployed and leveraged easily and all around, with or without the need to have a huge computing beast working in the mother-ship. It is an age where Edge Distributed Computing (EDC) is catching many eyeballs and wallets.

So how does it work? And why does it work?

Deep waters – busy waters

In a recent discussion on supercomputing, Rohit Kochar, Founder and CEO of Bert Labs, elaborated, “Traditionally, supercomputers were seen as single devices having the ability to store, process, analyse and compute multiple functions. However, for us, supercomputing means leveraging the Bert Platform Solution’s distributed computing capabilities across the Bert Nova suite of software modules. Each module serves a specific function, extending this distributed computing ability to the Edge with Bert Titan and Bert Aksh Edge Computing Devices.”

He further emphasised that supercomputing begins with Bert Maximus IoT-Powered sensing and Data Capture devices, where they’ve enabled super-computation on embedded boards. “Here, millions of data points are processed at the initial level, including numeric, alpha-numeric, digital images, and thermal images,” he said.

“If data transmission is costly and in certain scenarios prohibitive, edge distributed computing is a viable and effective solution.”- Prof Arun K Tangirala, Department of Chemical Engineering, IIT Madras

Elaborating on the relevance of EDC for today’s enterprises, Kochar highlighted its advantages over supercomputing in certain contexts: “A company may use it for HVAC, enabling real-time monitoring of chilling temperature, airflow, and volumes. Sensors can be integrated into various equipment to gather data on flow, temperature, and pressure, which serve as training data for AI software. Processing such vast amounts of data requires significant computational power, which can be overwhelming for a single location or plant. The solution lies in distributing this computing ability.”

Prof Arun K Tangirala from the Department of Chemical Engineering at IIT Madras agrees that EDC is a beneficial model for factories and manufacturing data intelligence, especially for all large-scale industries and factories that integrate several small units to manufacture products.

“The advantages include decentralised computing, eliminating the need for significant computing resources, and quick availability of models for each unit. Moreover, in scenarios where data transmission is costly or prohibitive, EDC and Federated Machine Learning or FML offers a viable and effective solution,” he stated.

Kochar adds how various IoT devices, themselves, work as the first layer of computing: “They capture data from plants or HVAC equipment and the first level of processing is done by Bert Mini. The solutions also make the data comprehensible for humans.”

The beauty of the solutions, as Kochar adds, lies in how miniservices architecture takes care of various aspects of data and processing. “Real-time performance is tracked by one miniservice, data is captured and stored by another, the next level of processing is done by another one, analytics by the next one, and so on for AI computation, prediction and optimisation.”

“The integration of 5G and edge computing delivers latency of less than 1 millisecond, significantly enhancing speed and responsiveness for applications.”- Nitin Bansal, Managing Director, India Head – Networks, Market Area South East Asia, , Oceania & India, Ericsson

He adds “We are the only company in the world where AI creation, training and inference can be done on the device itself. There is no need to take it to a remote or Cloud server – although we have those options if needed. In today’s complex and dynamic ecosystems, computing power is better in a distributed way instead of being in one location. It also helps that we do not have one monolithic software architecture to perform all functions. It is broken down into modules which work cohesively for the desired outcomes. Incidentally, many functions need to be executed in parallel, and not sequentially.”

As per Nitin Bansal, Managing Director, India Head-Networks, Market Area South East Asia, Oceania and India at Ericsson. “The integration of 5G and edge computing delivers latency of less than 1 millisecond, significantly enhancing speed and responsiveness for applications like automation, AR, VR, and real-time content delivery. This improved internet experience extends benefits to both consumers and enterprises. The proximity of data processing and storage to the source enhances security and privacy.”

New revenue streams are generated through services such as cloud gaming, smart manufacturing, autonomous vehicles, smart grid, predictive maintenance, and remote asset monitoring in industries like oil and gas, Bansal adds. “Telecom operators can capitalise on this synergy by offering edge computing services, optimising content delivery networks, providing IoT solutions, and enhancing AR/VR experiences, creating diverse opportunities for revenue generation across industries.”

Barnacles: power, costs and reach?

Would distributed computing have its limitations of bandwidth and resources? “That is where our IP shines. It is a combination of hardware, circuit design, choice of SoCs, firmware and associated components. The trick is how the software is embedded and how the AI modules are made Edge-ready for real-time training on data fetched from edge devices. Everything comes together on our solutions. We have filed patents at hardware, software, application and integration levels.” Kochar assures.

“In today’s complex and dynamic ecosystems, computing power is better in a distributed way instead of being in one location.”- Rohit Kochar, Founder & CEO, Bert Labs

The success of edge computing entirely depends upon the comprehensive deployment of small-cell network architecture to ensure no latency drop, adds Bansal. “Improved street infrastructure not only accelerates the expansion of 5G coverage but will also enhance network reliability and capacity, crucial for delivering high-speed data and enabling emerging technologies.”

The challenges are that each edge device needs to be equipped with reasonable computing power and well maintained, Prof Tangirala points out. “Moreover, the heterogeneity of computing power and data across devices have to be factored in. Finally, deployers of EDC and FML have to pay attention to model poisoning, essentially attacks from hackers.”

An interesting corollary to the argument of resource-intensiveness of distributed computing is that currently many use cases, as shared by Bert Labs, are around sustainability. “Most of our clients want to reduce energy consumption, power and fuel usage at plants and control raw material costs across the asset lifecycle. IoT devices do the first level of processing efficiently.”

Federated ML takes EDC to the next level by fusing the models developed on each device at a central server and pushing the fused model back to the devices.

Edge: Is it around the corner?

As to how easy this model is, Rohit cites many examples. “Like how a big FMCG major has used this in its Mumbai HQ. The platform integrated well with different equipment and BMS set-up via standard communication protocols. Only a few changes are needed to be made at the software and application level. These can happen easily and without touching their IP.”

As Prof Tangirala echoes, with a proper security and edge device maintenance strategy, the advantages outweigh the potential disadvantages making EDC, federated ML, useful for large-scale plants and also for industries that fuse information from multiple sites. “Of course, the necessary infrastructure including a central server that maintains the library of models developed on each device, fuses such models and communicates with the devices would be required.”

According to Bansal, the amalgamation of 5G, MEC, and IoT, IIoT is anticipated to effectively address the challenges and requirements of industrial applications, presenting a compelling proposition for various industries. In both consumer and enterprise verticals, the amalgamation of 5G and edge enables IoT and IIoT applications to run with high reliability, low latency, flexibility, and security.

Telcos can now offer edge computing services, optimise content delivery networks, provide IoT solutions, and enhance AR/VR experiences.

The synergy between 5G and edge computing is expected to enable predictive analytics, optimise operational processes, and foster a secure operational environment by addressing digital security and privacy risks. The enhanced connectivity will contribute to increased productivity for field and remote workers, while also supporting the creation of innovative customer experiences. Real-time data insights, a key feature of this integration, will expedite decision-making. “The combined effect is predicted to introduce new products, revenue streams, innovative work methodologies, accelerated process automation, and reduced dependence on fixed connectivity, collectively enhancing overall operational efficiency,” he says.

This is an evolving field with active research at many leading AI labs, Prof Tangirala reckons. “Our research group at Wadhwani School of DSAI is currently working on developing algorithms to address efficient model updates on edge devices, data heterogeneity and data drifts. As to FML: Federated ML takes EDC to the next level by fusing the models developed on each device at a central server and pushing the fused model back to the devices; essentially it works on model transmission and aggregation, as against data transmission and centralised model development on large datasets.”

It is not just about distributed over centralised, but also about agility, cost-effectiveness and decision-enablers. Edge has to deliver that edge to turn truly super—without being wobbly, without being too jelly.

 By Pratima Harigunani

feedbackvnd@cybermedia.co.in