Tool for prediction of component behaviour in monitoring system

Sometimes in a huge IT architecture with a lot of components like physical / virtual servers / databases / LDAP servers / micro services etc. on-duty engineers can see some correlations between metrics in one component and another. Looks like a magic or … butterfly effect in Chaos Theory.

Actually, we can research this dependencies for particular IT architecture more deeper and Marat made this research. He got a huge historical dataset with 5 metrics of each included server (or worker).

  • CPU
  • memory usage
  • Read/Write IOPS
  • network latency
  • disk size

He applied several Neural Network algorithms and models

  • LSTM
  • RMSE

… and finally Marat developed a tool which can predict a behaviour of component (it depends on component dependencies in selected IT architecture) for the next 5 minutes! Of course, you can say – “so, 5 minutes are really small time window interval…” But, based on my production experience , I can say 5 minutes is really good interval to be proactive for HighLoading.

Leave a Reply