ServiceNav artificial intelligence boosted by the experts of the AI Institute of Grenoble
MIAI Grenoble Alpes is a multidisciplinary institute in Artificial Intelligence, based in Grenoble. This center of expertise is directed by Eric Gaussier, formerly head of the Grenoble Computer Laboratory (LIG).
In 2014, within the framework of a governmental project "Fond Unique Interministériel" - a program allowing to finance R&D projects for a short or medium term market launch by associating the skills of large companies, SMEs and laboratories, Coservit and LIG got closer and started a succession of common projects: in 6 years, 4 Artificial Intelligence topics could be treated, involving about 20 experts and 10 publications of articles in scientific journals!
We reveal the stakes of these collaborative projects of high expertise...
Beyond the scalability of ServiceNav, the research work with the LIG aimed at adding Artificial Intelligence to the supervision in order to answer the following uses
- Reduction of false positives and false negatives
The market reports a number of false positives ranging from 30 to 80% for a conventional monitoring solution.
For example: for an average of 50 incidents per day, 25 of them are not really incidents. This is due to misconfigured thresholds: either too high or too low.
This means that operators spend half of their time on alerts that are not incidents or that are not a priority: time spent on useless actions, loss of confidence in the supervision solution...
- Incident prediction
When a critical problem occurs, the operations teams find themselves in "firefighter" mode, focused on this emergency. Whether during the day or on call, the stress generated is enormous, and the pressure exerted by management or users is complicated to manage.
Anticipating incidents with enough anticipation is therefore a solution for an organized and serene work mode, allowing at the same time to obtain a better quality of service and thus higher availability rates of business applications.
- Identify the source of a problem as soon as possible - Root Cause Analysis
Getting to the source of a complex problem takes time. The objective is to reduce this time to a minimum through analysis and reasoning assisted by Artificial Intelligence.
Eric: "The LIG is one of the most important computer laboratories in France, working in particular on distributed computing and the processing of large volumes of data.
The LIG research teams are interested in real production data to test and prove their algorithms. Coservit with its ServiceNav SaaS monitoring platform and its real-time streams of millions of data was therefore the right candidate."
As Eric explains, "The new issues are focused on causal relationships and not only on correlations between objects. This approach is new, difficult and little approached at the LIG. The ServiceNav "root cause analysis" project has therefore enabled the LIG to move forward in this vast field of causality with real-life use cases."
- Scalability and robustness of the monitoring platform
As a result of these 4 years of research, Coservit now has a strong Big Data competence. Few editors on the market have succeeded in switching to these complex technologies.
- Increase the availability of critical applications
By injecting algorithms into this "big engine" that is Big Data, forecasting alerts in the near future (+2h, +15h...) or further into the future (+3 months, +15 months...) saves uptime and reduces the pressure on operating teams both during working hours and, above all, during on-call periods.
Rachid: "working on causality allows to simplify the analysis of complex systems by focusing on the essential alerts of root causes..."
- Improve RoI of supervisory systems through improved operator efficiency and automation
As Rachid explains, "ServiceNav is now able to reduce noise, i.e. reduce false positives without generating false negatives. Adjusting thresholds automatically is the key to achieving up to 70% of noise reduction without changing the current paradigm and organization of operators. Our recommender system fits the current user processes with little change management.
We are now working on behavioral to automatically detect abnormal system changes on a large data set."