Looking Beyond the Hype: The Practical Side of AI in the Data Center
by Drew Robb
Many predictions, these days, center upon Artificial Intelligence (AI). We are told AI will impact every aspect of society. All facets of our lives will be enriched by AI technology. And of course, AI will pervade each and every element within the data center.
This may be true – eventually. But please note that Spielberg’s “AI” movie came out in 2001. Despite the AI hype, not much has changed in that time. And speaking of 2001, the Kubrick film of that title was released in 1968. Fifty years later, where’s HAL? The best we have is Alexa being able to tell us the weather or play a few tunes.
So let’s get practical about AI in the data center. What kind of tangible impact is it having NOW in terms of storage, applications and security that a data center manager needs to know about? In other words, let’s not worry about future potential – how can AI help immediately in the data center? And what should the data center manager be doing about it?
Be sure to check out the Data Center World session, "Artificial Intelligence and Deep Learning Coming to Your Data Center," presented by Remi Duquette, global head of data center clarity and machine learning for MAYA HTT, Ltd., on March 13 in San Antonio.
AI is Coming
Market research firm Tractica said the global AI market was worth $2.42 billion U.S. dollars in 2017. By 2025, AI is expected to be a $37 billion a year industry. That makes it something the data center manager better pay attention to.
“AI technologies are no longer science fiction, so data center managers need to prepare for them,” said Shariq Mansoor, Founder and CTO, Aera Technology. “AI is taking off at a rapid pace; it can help improve data center operations and services.”
The bad news is that it AI is excepted to eliminate 1.8 million jobs by 2020, according to Gartner. The good news, though, is that it will create about 2.3 million jobs over the same period. The interference here is that those making the right choices about AI today are more likely to be in a job in a few years than those who ignore it.
More good news comes in the fact that the largest portion of the AI market is enterprise applications such as image recognition, object identification, detection, and classification, as well as automated geophysical feature detection. AI requires complex, data-driven applications and those are in high demand in fields such as retail, healthcare, and automotive.
“Start planning for massive data storage capacity and scalability; demand for more elastic computing power including GPUs for AI workloads; and a new technology stack including open source technologies like Apache Spark,” said Mansoor. “From running autonomous operations, saving on power, performing predictive maintenance and continuous workload adjustments, AI is becoming a necessity to staying in business. Without AI, it will become almost impossible to run profitable data center operations.”
Said Tabet, lead technologist for AI strategy at Dell EMC, concurs. He said data center managers should be looking to harness AI to find better ways to optimize data center infrastructure. This includes leveraging sensor and related data to reduce power consumption, minimize downtime and detect anomalies as early as possible.
“AI will help data center infrastructure providers to deliver smarter infrastructure and connected assets to monitor, optimize and improve operations,” said Tabet. “These will include storage, compute and networking.”
He thinks we are at an inflection point with regards to AI. He offered some specific areas that data center managers should become interested in:
Automated conversation systems: More than simple chatbots, these are now capable of creating better customer interactions and user experiences. These will find their way into customer service apps, the help desk, and other apps aimed at improved IT resource and services. AI-based analytics will provide smart troubleshooting and diagnostic tools that can be used by the data center for problem resolution, proactive insight trend analysis, forecasting and resource scheduling.
Machine learning: Machine learning algorithms can be incorporated into the control layer of storage systems to enable easier monitoring of the various causes of traffic congestion. This allows them to predict potentially vulnerable sectors. Deep learning is part of a broader family of machine learning methods based on learning from data rather than harnessing task-specific algorithms.
“User requests and data traffic can be channeled to and from alternative storage locations based on network usage patterns,” said Shiladitya Chaterji, an AI analyst at MarketsAndMarkets. “Deep learning is an AI technology that can help to optimize infrastructure and operations, create more efficiency, and provide smarter predictive maintenance and related services that ultimately reduce costs.”
AI-powered infrastructure: AI is directly contributing to a more robust data center infrastructure, by incorporating GPUs and other accelerator hardware, such as AI-based appliances. Using AI to create smart infrastructure will help deliver a much more efficient data center, optimize configuration and enable better workload execution with dynamic settings and adaptive capabilities.
A little further into the future, natural language processing can support conversational AI. But that technology is still at the research level. The integration of advanced ‘agents’ and human operators is currently the better model versus a full automated approach, suggested Tabet. Advanced self-healing data centers with ‘autonomous’ capabilities are also further out.
Returning to the immediately practical, very large datasets are integral to AI. Data center managers should grow more accustomed to them. Data will come from multiple sources and is expected to require efforts to prepare, label and process. AI and machine learning provide techniques that can be applied to eliminate time and manual drudgery.
“A couple of practical AI areas for data center managers include policy based automation of routine tasks,” said Greg Schulz, an analyst with StorageIO Group. “This can encompass resource and service provisioning, the help desk, problem resolution, and active knowledge bases that capture new events, scenarios, symptoms and resolutions to help learn from the past to prevent the same issues in the future.”
Tabet believes that a changing landscape and architecture is coming for data centers. They will become more distributed and more compute will go to the edge or near edge. AI will be needed to deal with the complexity, data synchronization and analysis. But the needs of AI and machine data are quite different from other kinds of data. For example, machine data will require an immediacy of processing at the edge and a scalable, shared repository at the core. So the type of storage that may have traditionally been deployed may not fit a machine learning environment.
“Investment in AI or machine learning demands some strategic thinking about the underlying storage infrastructure,” said Matt Kixmoeller, vice president of product and solutions marketing at Pure Storage. “As they are data-intensive, and reliant on discerning immediate value from that data, solutions must be scalable and cost-effective, but also able to handle enormous data sets at high speed.”
He suggested the combination of on-prem and cloud-based storage solutions. On-prem data center elements will be needed for performance and predictability of cost. The cloud is there to quickly scale up and down in dev/test environments.
In this changing data center landscape, overhead becomes crucial. Expect a lot of overhead to be consumed by number crunching, analytics and data transportation. Depending on the architecture, this might be local or could span many systems and multiple sites.
This is where object storage is likely to come into its own for those investing in AI or machine learning.
“Don’t look at object storage as the ‘cheap and deep,’ but rather as the center of your differentiator in the future,” said Michael Tso, CEO, Cloudian. “The data center world is shifting and the winners will keep data in a format that is AI friendly.”
The use of AI in security is inevitable. After all, the number of new malware strains and viruses being created is staggering. An email security gap analysis found that out of a survey of millions of emails, 10.5 percent of this traffic contained spam or malicious messages missed by existing security tools. A good portion of this was spam. But the fact that about 0.3% were phishing messages and 0.04% contained malware attachments shows the trouble with modern-day security – no matter how good your tools are, they can’t catch all the nasty stuff. AI is needed to subject this traffic to more detailed and rapid analysis and to alert about potentially harmful actions going on behind the firewall – such as a unusual traffic patterns, ports being used in a suspect manner or data being transmitted externally.
“Some other applied and practical uses for AI include, security intrusion detection, access pattern learning of what is normal, abnormal, spam and malware detect, protect and prevent,” said Schulz.
Robots or Basic Automation?
No doubt somebody somewhere is pitching an executive about the wonders of robotic IT staff or a data center completely run by HAL or IBM Watson. Whether such wonders will ever come to fruition, time will tell. A better approach is to gravitate toward the tangible, obvious implementations.
“For now, focus on the low hanging fruit that bring practical benefit including chat bots, trend analytics, and simple automation,” said Schulz.