Machine Learning driven Closed Loop Automation

June 12, 2019
By Benoit Lourdelet

The agility at which the business can respond to real-life situations is proportional to the level of digitization that has been implemented in the business. For a business to nimble and agile, it is imperative that all the processes be delivered as a digital service that can be provisioned, monitored and remediated by an automation logic at the core of the business. The automation logic should follow the Monitor, Assess, Plan and Execute (MAPE) loop methodology to provide a complete closed-loop automation system for the digital business. Click below to see a video demonstrating using StackStorm with Machine Learning to do this.

In the video below, we demonstrate how StackStorm can leverage Machine Learning to drive a Closed-loop Automation workflow.

StackStorm is an automation system that can provision services based on process workflows created by the business owner. It translates Service specifications into atomic, traceable and auditable actions. Using sensors for the external environment makes it useful for keeping systems in the desired state by establishing a feedback loop.  The sensors, rules and actions enable StackStorm to deliver a MAPE loop automation system to provide closed-loop automation.

Achieving full system autonomy will up the ante of automation needed by a couple more notches. One must depart from naive system sensing and rely on in-depth insights to make sure the triggered corrective actions will augment the service rather than being detrimental to the cause. Machine Learning based system insights will provide the impetus to the Assess and Plan logics of the MAPE loop to accomplish this. The actions modifying the system are triggered as the result of the in-depth comprehension of the inherent system. However, these actions affecting the system are still defined and implemented statically. The logical evolution towards a completely autonomous system is to make the system modifications dynamic. While static remediations are necessary first steps, they are still rudimentary and naïve. 

The assumption that a human can craft a response to perform remediations in real time is being presumptuous of human capabilities. StackStorm augmented by Machine Learning algorithms will have the ability to not only trace actions sequences but to trigger the appropriate response, at the right time to achieve the results.

IT systems under supervision – take the example of a single Linux Host – are always complex entities. Sensors come to the rescue by factoring in the system response to the remediations. The remediation workflow is tuned based on the system measurement input. This dynamic control loop makes possible to optimize how the Automation is reacting to system disruptions.