DevOps and More Better Analytics
This post via Jenny Yang is by Toufic Boubez.
The main premise of the talk and the Austin session was that current monitoring tools are clearly reaching the limit of their capabilities. That’s because they are making some fundamental assumptions that are no longer true. Mainly, they are assuming that the underlying systems being monitored are relatively static and therefore their behavioural limits can be defined and surrounded by static rules and thresholds. It has become clear however that we have moved beyond static monitoring and that interest in applying analytics and machine learning to detect anomalies in dynamic web environments is gaining steam. However, understanding which algorithms should be used to identify and predict anomalies accurately within all that data we generate is not so easy.
I spent time talking about the importance of knowing your data and its characteristics in order to be able to use the appropriate analytics techniques. For example, techniques such as the three-sigma rule or the Grubbs score (check out kale, the most excellent tool introduced by @abestanway at Etsy) are only meaningful if your data has a normal probability distribution. I also covered the importance of context, and described some simple data transformations that can give you powerful results, such as the simple act of looking at a histogram of your data, because the regular time series plots can only give you so much insight.
I ended up straying a bit from my original topic, mostly due to a really good Open Space the day before about self-healing systems. I’d given a talk about control theory at the Bay LISA meetup a couple of nights before, and the combination of the excitement at each was too contagious to resist, so we got into a discussion about open and closed loop systems, which I had mentioned in a previous blog post. Finally, I broached the seemingly taboo topic of how much data to you really need? Of course, if you know me, this led to the Nyquist-Shannon sampling theorem . Oh yeah, there was also mention of cats and Disaster Girl.
Check it out. The video and slides are embedded above. Please feel free to send questions or comments. I might even reply! Next blog, I’ll look at some more analytics algorithms. Yes, I promise (this is reminiscent of the “free beer tomorrow” sign I see at some pubs).