The Development Of Artificial Intelligence And Machine Learning To The Edge Brings Opportunities For MTDC And High-density Wiring Solutions

- Apr 13, 2020-


Artificial intelligence and machine learning are evolving and bringing benefits and efficiencies to adopters.To that end, corning has teamed up with dallas-based Citadel Analytics, which has been deploying AI platforms in managed data centers (MTDC).

Image. The PNG

There is a common saying on the Internet: "knowledge tells you to put all your eggs in one basket. Wisdom tells you not to put all your eggs in one basket."Machine learning (ML) leads us to recognize the eggs and put them in the basket, but artificial intelligence (AI) suggests that we should not put them in a basket.

Jokes aside, AI and ML are used for everything from language translation to the diagnosis of complex diseases, far beyond vision.To give you an idea of how much computing power AI and ML require, baidu in 2017 demonstrated a Chinese speech recognition model that requires not only 4 terabytes of training data, but 200 billion billion calculations over the entire training cycle.

We need to strike a balance between satisfying the AI and ML requirements and providing the highest quality of service at the lowest cost.So how to provide the highest quality of service?By reducing the physical distance that data travels between the terminal device and the processor, the effect of delay on transmission is improved.We can optimize transmission costs and quality of service by building edge data centers closer to where data is created and used.The second is to seek the lowest cost.The cost of transmission increases with the amount of data transferred, the distance, or the number of hops.AI and ML greatly increase the amount of data transferred, resulting in higher transmission costs.Edge data centers are close to where the data is created, so they are increasingly important solutions, and a large portion of edge computing is deployed in MTDC.MTDC can provide the lowest risk of local data center deployment and can realize the benefits the fastest.

What is AI, ML, MTDC?

Before discussing AI, ML, edge data centers, and MTDC, it's important to take a closer look at their concepts to make sure everyone is on the same level of understanding.

Artificial intelligence is an extension of the theory of computer systems, capable of performing intelligent tasks normally performed by humans, such as visual perception, speech recognition, decision-making and translation between languages.If toy dolls are used to describe these relationships, then AI is the largest nesting doll, with machine learning in the middle and deep learning in the machine learning nesting doll.Machine learning is an application of AI that enables systems to learn automatically and improve from experience without programming.

Edge data centers bring the computing and processing power of the data center closer to where the data is created by dispersing some of the more delay-sensitive applications from the core data center.Multi-tenant data centers (MTDC) are also called managed data centers, and users can rent space to host their facilities.MTDC provides space and networking equipment to connect users to service providers at minimal cost.Users can rent space, server racks, or complete dedicated modules to meet their various needs.

Image. The PNG

How good is AI?

AI and ML are the most revolutionary technologies we have seen since the advent of electricity.It is more powerful than the Internet and mobile revolutions combined.What makes AI technologies so powerful and influential is that they can understand large amounts of data quickly and efficiently.We live in a world where data is constantly being generated and driven by data (market analysts predict that more than 80 per cent of the data that exists today was created in the last two years) and without the tools to understand it, we will be swamped.

To take a simple example, the world will create about 40 megabytes of information this year.That's 40 trillion gigabytes of information.Humans cannot understand all this information, and even if everyone worked around the clock, it would be theoretically impossible.

So how do we make sense of all this data?The answer is to use AI and ML.These technologies prefer data, which is like their oxygen.By using a powerful and properly trained AI/ML model, we can accurately process large amounts of information, revealing very valuable data to guide our actions.

The ML model of MRI is a good example.They tested for known or non-cancer results (positive and negative), a process known as training.A new set of MRI was then loaded into the trained model for analysis.These new mris are called validation data sets.The data is run through a trained model, the results are calculated and displayed, and the results are then evaluated against the performance metrics selected for the model.If the results are acceptable, the model is trained and more testing and validation are prepared.If the validation data does not meet the metrics, we go back and either redesign the model or provide it with more data to provide better training for the next validation test.This phase is called the validation phase.

Image. The PNG

The benefits of artificial intelligence sometimes come with an extra twist.For example, most of the businesses Citadel Analytics deals with are businesses where customers expect AI technology to improve efficiency in order to generate sales growth or lower costs.However, they soon found that AI could greatly improve the work efficiency and pleasure of employees, and the biggest beneficiary was the employees.A company that knows how to use AI/ML also tends to have employees who are more satisfied with their jobs than a company that doesn't use AI.

This makes sense because AI/ML is all about automating "boring" things and getting your people to do what they are good at and passionate about without reducing their productivity.The great benefit of using AI technology is to make employees happier and reduce employee turnover.But this is something that many companies don't take seriously at first.

How to deploy AI/ML and provide a reasonable wiring solution

AI requires huge processing power, which has always been a problem.Thankfully, companies like NVIDIA, Intel, and AMD are closing the processing power gap.This allows companies such as BMW, wal-mart, Target and more to deploy edge AI capabilities.This requires the installation of powerful hardware that will use a pre-trained model to process local data.This greatly reduces wait times and the need for real-time bandwidth.

The problem is that no one can really do edge AI/ML deployments, because while the hardware's pre-trained model can handle data, it doesn't have the capability to update the model and make it more powerful.

As a result, hybrid solutions were developed.

In a hybrid design, the edge server will use a trained model to process all local data.These "best servers" may be located in different MTDCS, allowing for flexibility in network and application selection.MTDC's optical network infrastructure is typically deployed using single-mode fiber to meet end-user requirements for future expansion.For companies deploying AI/ML, it is important to consider both current and future network bandwidth requirements.Citadel Analytics typically has a rule of thumb: take the average of the expected bandwidth and multiply it by 4, which is the amount of bandwidth that should be deployed on its system.In current AI/ML deployments, bandwidth is particularly important and a primary consideration.

The increased bandwidth also highlights the need for high density solutions for MTDC and end users.The high density scheme maximizes the benefits of MTDC space, and the end users can use their invested space more efficiently.Because the infrastructure varies with the needs of different users, it is important to find product vendors with product widths (single or multi-mode, LC or MTP, etc.) and extended flexibility.One way to reduce costs for end users is to increase density and reduce power consumption.This can be achieved by parallel optical devices and port separation, that is, by using branch cables to decompose a high-speed bandwidth port into several low-bandwidth ports.For example, a 40g parallel transceiver port is split into four separate 10g transceiver ports for high-density cable conversion.

Delays can typically be reduced by 45% by reducing transmission distances within 10km.What does that mean?This will drive the deployment of more and smaller regional data centers closer to where the data is generated.MTDC will be the primary hosting facility for these smaller data centers.The edge data centers will be hosted in these interconnect intensive MTDCS, which will complement each other to provide a more comprehensive service to customers.MTDC, with its interconnected infrastructure, ecosystem and rich customer portfolio, will be able to capture the business opportunities of edge computing development.

expect

AI and ML technologies will continue to generate revenue and productivity for adopters.We will see more companies running applications on edge MTDC in the future.For these companies, MTDC offers lower risk and faster revenue growth.This will lead to the emergence of more marginal regional data centers, rather than the centralized large data centers we have seen before, as well as more interconnected facilities and higher density solutions.