Why static thresholds just don't work

Earlier this summer, I had the honour and pleasure to join AIMS CCO Kristina Mordal in meetings with several developers and IT managers when she visited the Netherlands. I saw her educate these potential customers and newly-generated AIMS fans in a way that made it absolutely clear why setting static thresholds is not good enough.

Inspired by her talks, I took a picture of her drawing and remembered her words, so I could share them in a blog post.

Monitoring using machine learning

AIMS uses artificial intelligence to monitor SQL, BizTalk and OS, using many different metrics in advanced algorithms to decide what behaviour is normal for your environment.

The algorithm combines metrics to decide the normal message volume precisely for any particular time.

Imagine this is your normal messaging volume in a month, and based on what AIMS has learned about your environment it can set an exact line representing the volume.

When setting a static threshold, you will have to take into account the highest and the lowest message volume in a month. This is represented in the drawing below.

For this month, you will be warned when at any given point the message volume is higher than 8500 or lower than 3500. 

This means, that when there is a deviation in message volume that falls between these lines, you will not be alerted.

For example, if the message volume on the 6th of the month is 6500 instead of the normal 4000, you would not be warned.

Same applies when on the 14th, the message volume is only 6000 in stead of the expected 8000.

AIMS detects normal behaviour patterns, knowing exactly when the system will receive what message volume. It knows the accepted deviations and understands when it is considered an anomaly. This way, you will never miss a warning, and you will only receive one when it is definitely to be considered abnormal behaviour. With AIMS your monitoring tool can detect this, based on deep knowledge of your environment through artificial intelligence and machine learning algorithms. 

You've got to love science!