Saturday, February 2, 2019

AIOps Now Shows how AI is Applied to Network Operations

I just got back from chairing the new AIOps Expo, sponsored by TMCnet, a three-day event looking at the advantages and challenges of artificial reality as applied to management of information technology systems (both enterprise and communications service provider).

These days, it always is clear that value and opportunity for connectivity providers largely has shifted up the stack to applications and platforms.

One big question asked by enterprise buyers at the event is how “real” AIOps is, where and how it can be used today, and what the roadmap looks like.

As with all new buzzwords and trends, we have to define what we are talking about. According to Gartner researchers, “AIOps platforms combine big data and machine learning functionality to support all primary IT operations functions through the scalable ingestion and analysis of the ever-increasing volume, variety and velocity of data generated by IT.”

Putting the AIOps moniker into context, one might argue it is the latest way to describe the use of artificial intelligence (machine learning and deep learning more than neural networking at this point) to improve IT operations.

You can get an argument about where “automation” ends and “AI” begins; that was clear from discussions at the event. You will not get much argument that, at the present state of the art, it is much easier to apply AI to specific functions and processes than to integrate and correlate all functions and processes.

Compared to a decade ago, when the ability to analyze “big data” was the buzzword, insight still is the desired outcome. What seems different now is that Moore’s Law makes a difference.

Analysis that might have been just as valuable 10 years ago--or two decades ago--now is feasible because the costs of computation and storage as so much less.

But one new emphasis is on machine learning: allowing AI to work autonomously--with human approval--to modify system behavior based on what has been learned, without human intervention.

One concrete difference is the role of scripting and code writing. Compared to present practice, the goal is to allow machines to modify their own behavior without direct coding labor. That obviously raises clear issues about bias in the coding systems and security and privacy issues.

But a big strategic change is the shift to allowing machines to discover patterns that would be prohibitively expensive we were to attempt to discover patterns using human agents only.

One illustration of the potential benefit can be glimpsed if you ask why network operations centers have so many screens, as technologist Frank Yue, KEMP Technologies solution architect. The reason is that each subsystem has its own management and monitoring system.

That means fault isolation is more complex than it would be if all systems were correlated, if all the data could be analyzed and understood in ways that reduce the total number of alarms, for example. The cascading alerts NOCs have to deal with is itself a problem, as many of the alerts from different systems actually refer to common events, noted Bhanu Singh, OpsRamp VP.

Frank Yue, Kemp Technologies

In fact, as noted by Taly Dunevich, Ayehu global VP, AIOps is, in many ways, AIOps is the latest way to automate IT processes, without scripting or coding.

All of which lead Wayne Parker, Northrop Grumman technical advisor to quip that “in 10 years we’ll be calling it something else.” Indeed.

No comments:

Public Policy is Devilishly Hard Stuff

Public policy success always is harder than you might think, if only because the causal relationships between a policy and an intended out...