Metafor Software: Using Machine Learning To Predict Server Issues

Max ‘Beast from the East’ Smolaks covers open source, public sector, startups and technology of the future at TechWeekEurope.

If you find him looking lost on the streets of London, feed him coffee and sugar.

Follow on: Google +

Toufic Boubez, co-founder of Layer 7, wants to pre-empt problems in the data centre with the help of clever algorithms

Dr Toufic Boubez is the co-founder and former CTO of Layer 7, a small Application Programming Interface (API) security company which was recently sold to CA Technologies for an undisclosed but “very generous” amount.

The acquisition was announced less than a week after Intel agreed to buy API expert Mashery, a 125 person firm, for a reported $180 million, leaving analysts to ponder the sudden land-grab in the API business.

Prior to co-founding Layer 7, Boubez was the chief Web services architect for IBM’s Software Group and drove its early XML and Web services strategies.

His new venture, Metafor Software, is designing a set of software tools that automatically detect unexpected changes and hidden anomalies in the server room. TechWeekEurope asked the Canadian entrepreneur to explain why he chose to work in this relatively unknown field.

Solving future problems, today

Boubez, an expert on Service-Oriented Architecture (SOA) who contributed to several relevant specifications and standards, essentially built the Layer 7 platform from scratch. When he started in 2002, APIs were still a new and unexplored topic.

“I’ve spent the first couple of years ‘evangelising’, telling people about this new way to build software, as opposed to actually selling stuff. It’s hard to be a start-up with the product that you have to evangelise, as opposed to just riding the wave,” told us Boubez.

metafor logo BIG (Small)Now, when the industry has finally embraced APIs, he has co-founded another business based on an innovative, untested idea – Metafor Software. “What I noticed in the last few years was this trend towards virtualisation and cloud. But I also noticed there was a lot of talk about DevOps. And I thought that many customers that were using and installing our product [Layer 7] in the data centre were totally unequipped for this revolution.”

“If you have a big enough credit card you can spin up a whole data centre in an hour. But then, on the software deployment side you’ve got APIs going up and down, servers and infrastructure changing all the time, creating all kinds of chaos in an organisation. That got me thinking about what the next big thing is going to be.”

And the next big thing, according to Boubez, lies in the field of “anomaly detection”. This conviction has led him to set up Vancouver-based Metafor Software, which enjoys full support and participation of his colleagues from Layer 7. Anomaly detection is a particular category of data centre tools that highlights events that fall outside of the usual patterns, helping to nip any potential errors in the bud.

Server Anomaly“If you have a cluster of servers running MySQL or Apache, they should behave like perfect clones of each other. But you would be surprised how frequently they get out of sync. When you are updating them, a handful won’t get updated properly or get the wrong package, or somebody goes in and makes a change to fix a customer problem, and suddenly there are outages, there is downtime,” explains Boubez.

“We are building software that watches these servers and detects every instance when they go out of sync both from the configuration perspective, which is static, and the behaviour perspective, which is dynamic. In the end, it saves time and money.”

To achieve this goal, Metafor is relying on clever algorithms and machine learning. The response of its software will depend on how much information is available. Boubez says that, as the company processes more data, its software will learn how to make better decisions.

“Security people have successfully used machine learning for many years. You look at intrusion detection, threat detection methods, all of it is based on machine learning. But the data centre hasn’t adapted these methods, so that’s what we are trying to introduce. We have the resources now, to apply serious algorithms to large data problems.”

Right now, Metafor Software is tiny – a team of seven engineers and one marketing expert, plus a few PhD interns working on machine learning. But it has big ambitions: Boubez wants to build a product that can not just alert about anomalies, but also help normalise them with a single mouse click. Sure, it sounds like a tall order: but it is worth remembering that Layer 7, now part of one of  the biggest independent software corporations in the world, also had humble beginnings.

Are you a patent expert? Take our quiz!