Adding a Human Element to ML-Powered Server Administration

The complexity of today’s virtualized IT environments is a big concern to organizations that employ administrators to manage servers and the databases and applications that run on them. It’s what drove SIOS Technology to adopt machine learning technology as a path toward IT simplification. Today the company announced that it’s tapping the knowledge of experienced managers to help train the algorithms.

Earlier this year, SIOS started selling SIOS iQ, which currently only supports VMware‘s ESX Server hypervisor. The software employs machine learning and graph analytic methods to detect anomalous behavior within virtual machines (VMs) and to shorten the amount of time spent troubleshooting problems.

The analytic horsepower is necessary, the company says, due to the increased level of abstraction between VMs and the underlying hardware resources (CPU, memory, storage, and network I/O) they consume. Trying to track the impact of so many variables over time in environments that may contain more than 1,000 VMs across development, test, production, and high availability environments is simply too difficult for mere mortals.

SIOS’s end goal is to give administrators a leg up on tracking what’s “normal” behavior in the uber-complicated virtual environments, and alerting them when something goes awry. After just seven days of watching how a particular system works, the algorithms can step in and start in and become a first line of defense for the administrators.

With the launch of SIOS iQ 3.3 this month, the company is augmenting the unsupervised machine learning algorithms with some human-curated data. According to SIOS Director of Product Management Jim Shocrylas, it’s all about boosting the effectiveness of the algorithms.

“We’re leveraging human knowledge to supplement machine learning knowledge,” he tells Datanami. “Maybe the administrators know that we’re a little too sensitive to the performance of their infrastructure, so they can turn some knobs to detune a little bit so that it better represents what their expectations are in terms of performance.”

Servers are built by humans, and they sometimes reflect human quirks and foibles. An experienced IT administrators who has put his hands on a given machine will often know about performance intricacies, and this information can be used to speed up the training of the algorithms.

“There are situations in the pattern that we recognize where the IT guy says ‘This is not an issue you need to worry about.’ So I can incorporate that into our learning segment,” says SIOS CEO Jerry Melnick.

The algorithms are still steering the ship and in charge of detecting anomalies in the long-haul. But according to Melnick, the combination of human intelligence and machine learning will help SIOS customers get to their destination more quickly.

“Think about it like a self-driving car, where we allow you to hold the wheel every once in a while,” Melnick says. “Quite frankly, being able to tell it what patterns are relevant and which are not is probably the most important semi-supervised learning aspect of the machine.”

Eventually, the time will come when algorithms will guide the whole shebang – from configuring VMs for certain jobs (dev, test, production, HA) and then manage themselves too. Longtime servers watchers may recall how vendors like IBM touted the rise of so-called “autonomous computing” more than a decade ago.

The technology wasn’t really ready back then—Watson was but a gleam in the RS/6000 division’s eye. But thanks to the advances in the application of machine learning technology over the past decade, the prospect of machine-governed machines is getting much closer to reality.

Related Items:

Corralling VM Complexity with Machine Learning

How Machine Learning Is Eating the Software World

(feature art: Bluebay/Shutterstock.com)