Network operations: A new role for AI and ML

Artificial intelligence (AI) and machine learning (ML) are still viewed with skepticism by many in IT, despite a decades-long long history, continuing advances within academia and industry, and numerous successful applications. But it’s not hard to understand why: The very concept of an algorithm running on a digital computer being able to duplicate and even improve upon the knowledge and judgement of a highly-experienced professional – and, via machine learning, improve these results over time – still sounds at the very least a bit off in the future. And yet, thanks to advances in AI/ML algorithms and significant gains in processor and storage performance and especially the price/performance of solutions available today, AI and ML are already hard at work in network operations, as we’ll explore below.

The primary motivations for the adoption of AI and ML in day-to-day operations include the increasing complexity of network solutions, most notably on the wireless side; lack of a sufficient number of network professionals to handle the increasing scope and scale of network operations; ever-present requirements to minimize labor-intensive operating expenses; and continuing efforts to increase end-user productivity and assure the network capacity essential to growing numbers of end users with multiple mobile devices simultaneously in use, especially running time-bounded applications.

Essential limitations on human performance is another factor; it’s increasingly unreasonable to assume that even the best operations professionals can simultaneously consider the number of variables present in today’s networks, especially while keeping up with new technologies and products. As a result, embodying smarts in AI/ML-based products and cloud-based services is rapidly becoming a front-burner interest for even the skeptical.

Defining artificial intelligence and machine learning

AI and ML, while still continuing to evolve, are in fact relatively mature technologies, with production deployments going back to the 1980s. Simply put, AI is the emulation of human knowledge captured and engineered into algorithms and operational solutions often called expert systems. ML is the ability of these algorithms to improve their performance, based on operational experience, but without manual intervention or conventional reprogramming (but, of course, often via feedback from human operations professionals). Such technologies as neural networks and deep learning are often applied; consider IBM’s Watson solution, which has demonstrated benefits across multiple applications.

Without AI, such capabilities as modern control systems (including those within commercial aircraft and similar mission-critical settings), healthcare, financial markets, and many more simply could not function with the reliability and availability that is essential to success. This last point is critical – whereas humans can never be 100% productive, AI/ML solutions can chug away on a 24/7/365 and even global basis, simultaneously considering and acting on a vast array of variables that would be beyond the grasp of even the very best human experts in any field.

To quantify the benefits of AI and ML, we sought out end-user and service-provider organizations that are already making use of AI/ML-based network-operations solutions today. Their experiences reveal the operational requirements and challenges that are being addressed, what benefits are being realized, and what’s on the wish-lists of these early adopters.

MSP multiplies staff capabilities with AI

Technology Engineering Group, based in Medina, Ohio, is a full-service IT and network reseller and consulting firm. Among the wireless-LAN product lines they carry is Mist Systems, which has risen to prominence in the past few years in part thanks to its positioning as the “AI-driven WLAN.”

“We’re network architects,” says Jon Strong, managing partner and co-founder at the firm, which specializes in large networks, including wireless, for schools, businesses, municipalities, manufacturers, and office environments in northeast Ohio. He emphasizes the essential need for AI-driven analytics: “While cloud-based WLANs are clearly the trend, there’s still a need for improved analytics. Even for experienced technicians, it’s often hard to understand what’s not working – and situations that are difficult to diagnose can represent an enormous resource sink and productivity loss all around.”

What appeals to Strong about the Mist AI-based solution is that “it looks at network from the client up. Bottom-up troubleshooting is the most effective strategy, and Mist offers everything I’ve ever wished for.”

As an example, Strong notes his experience with the North Canton City Schools. “We needed better visibility into the operation there. We deployed 314 APs in just one month, resulting in better coverage, improved visibility, continuous proactive monitoring, and insights well in advance of issues becoming visible to users.”

At another school district, “we discovered a VLAN/DHCP problem that had been there forever, and Mist is still finding problems that would be difficult even for a highly-experienced network engineer,” Strong says. He also mentions Mist’s Marvis Virtual Network Assistant, another incarnation of AI which he notes “enables natural-language queries of even low-level network and client issues.”

Another benefit AI can provide is an abstract view of network, enhancing network tech productivity via a high-level, rather than an element-based, view of the network: “The data we need is available in a readily-usable form,” Strong says.

AI and ML enable productivity gains

Northgate Gonzalez Markets is a chain of specialty food stores with 40 locations in the Southern California area. The company operates two data centers, a 400,000 square-foot distribution center with many different temperature zones, and an associated financial services organization. Their WLAN includes about 500 APs from KodaCloud, a supplier that’s focused on AI-driven, cloud-based Wi-Fi solutions since their founding.

“We were intrigued by the possibilities of both a cloud-hosted Wi-Fi solution and AI and ML for network operations,” says Harrison Lewis, CIO of Northgate Gonzalez, which chose KodaCloud to replace its previous more-traditional solution. “Just for starters, when KodaCloud APs come up, they automatically collect information about environment, clients, and load, and self-configure, placing no demands on our operations team. We’ve also experienced automatic problem resolution – such as that related to signal coverage – exceeding both our goals and our expectations.”

The mission-critical nature of Northgate Gonzalez’s IT operations provided further incentive to look for an AI-based solution. “All of our processes rely on wireless, with the exception of back-office accounting and our support center. AI and ML are enabling automatic problem resolution, and, with the end of trouble tickets, we’ve seen a reduction in the demand for our technicians to near zero – a 100% gain in productivity,” he says.

Harrison also notes that AI has simplified the introduction of new client devices, and, with guest access for roughly 400,000 additional users now being piloted, “we don’t want to be in the situation of having to grow our support organization to meet this new demand.”

Harrison has high hopes for realizing additional benefits from other applications of AI across the organization, “in financial services, compliance, know your customer, fraud detection, human resources management, network security, data loss prevention, and more. The key justification for now is meeting the need to more intelligently identify network performance degradation and disruption, and automatically respond in an optimal fashion. But the benefits are far-reaching – dealing with transient loading issues, isolating problems related to class of service, enhancing reliability and continuity, and optimizing Cloud services, and, again, that’s just for starters.”

Uptime, performance get a boost with AI, ML

Faramarz Mahdavi is senior group director, IT infrastructure and operations, at Cadence Design Systems, a leader in electronic design automation. Cadence’s network regularly sees 8,000 users across 60 locations, with on the order of 1,500 APs, and with wireless being the primary access for most. Cadence is using both wired and wireless gear from Aruba Networks, and recently completed a major network upgrade at the San Jose, California, headquarters.

“We’re not yet major users of AI and ML solutions, but we see the value in exploring a number of directions here,” Faramarz says. “We have deployed a chatbot for basic self-service user help-desk functions and problem resolution. On the network side, we’re currently using Aruba’s Introspect for behavioral analysis, and we’re also looking at Aruba’s NetInsight, which can provide actionable recommendations for remediation, and ClientMatch for automatic RF optimization. The goal, of course, is to be more proactive and to leverage ML in order to recognize usage patterns, alert us to anomalies, and, ultimately, to provision automatic problem resolution. That’s the key – turn a reactive approach into one that’s proactive, preventing outages, and adjusting configurations well before users notify us of problems.”

“It’s important to have a solid infrastructure in place before deploying AI,” Faramarz says. “Supplier vision and product portfolio are also key – we want to make sure that AI and ML are deployed as extensions to what we already have working. But with uptime and performance – including that of time-bounded services – always the top goals, AI and ML are key directions for us. In fact, security is always our highest priority; AI and ML together, however, are the clear number two.”

Looking ahead: AI and ML in SDN, NFV

Network analytics, both wired and wireless, has over the past few years opened the door to the greater utilization of AI and ML in organizational settings. Analytics is the set of tools applied when one doesn’t know what one is looking for, dealing with the more-variables-than-equations nature of so many performance problems, security challenges, and otherwise poor operational behavior by extracting meaning and value from network logs, databases, and other voluminous source of information that would be challenging for mere-mortal human network professionals. With the value of analytics now proven, AI and ML are poised to complete the feedback loop between analytics and management consoles. This form of automation can help reduce costs while enhancing reliability, availability, overall performance, and the productivity of network-operations teams and end users alike.

AI and ML can also play a key role in the success of other networking technology initiatives, including SDN, NFV, and cloud-services integration. And as is usually the case with major steps forward, questions regarding reliability, applicability, cost/benefit justification, industry standards, and APIs are already hot topics – but that’s a good sign for the future value and success or AI and ML in network operations. It’s in fact becoming difficult to imagine a future for network operations where both of these technologies play anything less than a vital role.