What role does AI have to play in the modern data centre?
Article by Scale Computing co-founder Jason Collier
The prominent role given to AI (artificial intelligence) at the recent Microsoft Build conference by CEO Satya Nadella and at Google I/O by CEO Sundar Pichai, gave a very clear indication of just how important it is becoming.
According to a survey by Forrester Research last year, 70 percent of enterprises expect to implement AI in 2018 and 20 percent said they would deploy AI to make decisions and provide real-time instructions. With this in mind, it’s important organisations have the supporting infrastructure in place that will develop AI applications and provide the speed and performance needed. Organisations are adapting to new and modern requirements, and in turn placing new demands on the traditional data center.
However, today’s modern data center has evolved enabling organisations to meet new requirements with a scalable and flexible infrastructure that delivers performance and speed. Separate research by IDC predicts that global spending on cognitive and AI systems will grow nearly 55 percent this year to US$19.1bn and that 75 percent of enterprise applications will use AI by 2021. With this speed and rate of growth, data centers need to support and underpin the role of AI.
With retail, banking, discrete manufacturing and healthcare providers all predicted to spend significant sums on cognitive technology and AI, it’s clear AI is not just about creating a voice on a device. With the IT department of the future likely to include IoT devices, micro-data centers, cloud-based computing and more traditional data center components, organisations are looking to the data center to bring this together into a an overall IT infrastructure strategy.
However, AI is still in the early stages. There aren’t that many, if any, organisations that have come close to where Google is with AI in the data center. Google has used the DeepMind AI engine to make its data centers more efficient, by incorporating a system of neural networks. But to do this effectively, requires a firm grasp of the mechanics, extensive training and huge test sets to validate the data before it is ever put into production. To develop and utilise neural networks correctly, organisations need significant expertise and computing resources. Google has it but very few others do.
That’s not to say organisations can’t put themselves in a position to prepare for AI in the data center, so that when the time comes they are ready and prepared. However, there are a number of issues that organisations need to be aware of if they want to use AI effectively.Look beyond the hype
If organisations want to assess the benefits of AI properly, it’s important to look past the hype. It’s easy to underestimate the amount of time, knowledge and data required to implement AI systems effectively and there’s a real danger of handing over decision-making to AI too early in the implementation process. AI needs time to learn and to develop in the environment, before it can be trusted to make decisions and take actions.Get a grip on management
Organisations will increasingly need to rely on automation to keep pace with the growth in compute and the distributed nature of compute resources. However, this does not mean you need complex algorithms from neural nets to achieve increased efficiency. Effective data collectors that feed into a condition system are able to inject good data. Along with state machines that take action on changing and relevant conditions, this can provide a very effective step towards creating self-healing data centers.
AI can provide powerful capabilities but it is hard to exploit them if the team cannot manage the AI systems and glean insights from the information gathered.There’s only one Google
As mentioned, whatever objectives an organisation might have for AI in the data center, emulating Google should probably not be on its list. Google’s DeepMind AI engine incorporated a system of neural networks but developing and utilising those networks effectively requires a huge amount of expertise and computing resources. Resources that most organisations just don’t have.AI is not the answer to everything
AI is not going to solve every problem. It helps but it isn’t a magic bullet that can cure everything. So organisations should exercise caution when deploying an AI-driven service. There is no point-and-click, off-the-shelf AI software that “makes my data center work better.” Organisations that share this feeling will be set up for failure. It’s key to incorporate AI into facets of an organisation, but it is also important to utilise other key data center technologies alongside it.
Right now, there’s a lot of buzz about AI. But there are still plenty of issues that need to be addressed and overcome before it becomes an everyday reality in the data center. It’s not as clear and simple as some might lead people to believe. AI is not exactly stuttering, but there are still a lot of “ums” and “ahs” punctuating its progress.