Time Series Analysis – Predict Alerts & Events

Hello All,

This post is intended to give you an understanding of upcoming series of Timeseries analysis and how it can be useful to predict issues that may arise in future or predict growth etc.

Think of this, Many of the SRE's or admins work today mostly by doing reactive fixes (some may not) as they get alerted for something and they will fix it.

How about predicting those occurrences and adjust them proactive. isn't your oncall or day job is easy?

In my point of view (other may differ), most of our incidents, alerts that we receive are timestamped with there possible threshold breach value of that metric and they are exactly time series data as like other data sets like share price, weather, population. etc. And if you want to learn some machine learning or AI/ML models why cant we play with our own data and do a meaningful research or POC for productive result instead downloading datasets of irrelevant and do practices on them.

Here is it, as we figured out we have the time series data what all we can do.

Typically these are the alerts that we get as SRE's or Admins,

1. Service Down/up
2. Filesystem Alert
3. Storage Growth
4. High number of connections
5. High Load Averages
6. I/O Latencies
7. DB Response Time

Is it not? So what can we do with them.

Per say, If we have data of that particular server up/down for last 12 months or 30 days, we can predict and tell what is the probability that this server can go down or particular can down in next 30 days.

Or, if we have data of filesystem alerts of last 6 months, we can foresee when the filesystem reach 100 percent based on usage in past.

And, we can predict when is the database response time is going to be bad based on previous data points by gathering SQL response metric from Oracle database.

All we can do is to do Time series analysis and do prediction using Linear Regression models using python statsmodels, sclearnkit, fbprophet etc.

Hope you got the gist of what I am going to do in next few posts.

1. Intro - Time Series analysis Fundamentals
2. Time Series analysis using python statsmodel, ARIMA
3. Time Series analysis using fpprophet
4. Conclusion

Let me post what could be the out come of doing this. Ex: Database growth and filesystem growth predictions.