How Cisco-owned AppDynamics and Perspica bring machine learning to users
- 19 July, 2018 11:00
Application performance management (APM) vendor AppDynamics is bringing more machine learning into its platform thanks to a 2017 acquisition - here's how it did it.
On 1 November 2017, Cisco completed the acquisition of machine learning and data processing firm Perspica with the aim of merging its AI technology and expertise with its application performance management (APM) and IT operations software department AppDynamics, which it had acquired in March that year.
Now that strategy is starting to pay dividends as AppDynamics plans to bring machine learning features like prescriptive root cause analysis and predictive anomaly detection to its users this year.
Speaking to Computerworld UK over the phone Jean-François Huard, cofounder of Perspica and now CTO at AppDynamics explained how the acquisition was made for three key reasons.
First was Perspica's ability to infer anomalies from streaming data, giving nearer to real time insights than traditional analytics.
Then there was its focus on AI and machine learning specifically for IT operations and root cause analysis for application performance issues. Finally there was the team and its expertise.
Building AI into the AppDynamics platform
The early stages of the collaboration involved merging Perspica engineers with AppDynamics product teams. Then came the technical backend integration work, "focusing on those AI and ML capabilities to make AppDynamics a much stronger platform," as Huard put it.
The first feature users should be able to get their hands on this year is an automatic root cause analysis tool, which uses AI to spot anomalies before customers are impacted, similar to something rival vendor Dynatrace released last year.
"For end users it will look similar to how they get notifications today," Huard said. "They will get machine learned anomalies as events and deal with the events from there.
"With a double click they can get to root cause mode, where suspected root causes will be ranked for quick investigation. Otherwise it's the same product, the same flow and admin capabilities."
Now that the acquisition is starting to bear fruit Huard and his team are busy working on a roadmap to make AppDynamics the smartest APM tool on the market.
"So specific actionable recommendations on how to fix issues so user can be more effective in solving the problem," Huard said.
Going further, an AI-powered auto mitigation tool, where the AI will actually try and resolve an issue, is something AppDynamics customer Just Eat has been talking about wanting for a few years now.
Huard's team is also looking to make the AppDynamics BusinessIQ tool smarter to make forecasts on business data, like sales or short term revenue.
They also want to find 'cross domain event correlation', so helping customers understand if a slowdown in business activity is down to poor IT infrastructure, a bad user experience or even if a marketing campaign is not living up to expectations.
Machine learning is clearly the key battleground for vendors in this space, with the smartest platform delivering the most value to users that are data-rich and insight poor.
The scale of modern applications and consumer expectations when it comes to downtime means a more automated approach to APM, where operations teams can get ahead of issues, is vital.
Machine data specialists Splunk has been looking to integrate machine learning powered insights into its tools since 2016.
It now boasts advanced capabilities across its products, including its APM tools Splunk Enterprise and Splunk Cloud, as well as IT Service Intelligence (ITSI), User Behaviour Analytics (UBA) and of course the Splunk Machine Learning Toolkit.
Rival APM vendor Dynatrace is going even further, announcing its intention to shift away from app management to focus on 'software intelligence' earlier this year.
John Van Siclen, CEO at Dynatrace said: "Not only is digital transformation cloud-first, but as it relates to us, the cloud has changed everything because all the monitoring that we used in the past is now... it actually doesn’t work, it doesn’t deal with scale of the cloud, it doesn’t deal with the dynamism of the cloud or the new application models."
(Reporting by Scott Carey, Computerworld UK)