AIOPS

What to look for when selecting a AIOPS partner / Application

Written by: Marcel Koert B.S.E.E. | Posted on: | Category:

Introduction

In two of my previous blogs I talked about what AIOPS can do for you. Now I would like to talk to you about what AIOPS tooling needs to have to be as useful as possible.

Requirements for the AIOPS tools

• Ingesting data from multiple sources. • Analytics on real time data on the moment of ingestion. • Historical analysis of stored data. • Provide access to the data. • Storing the acquired data. • Use machine learning to analyse the data. • Being able to take action on result of analysis. • Large set of integration options.

Questions & Considerations

AIOPS

Ingesting data from multiple sources.

You will need to look at your data creating streams. Where is your data located and can the tool ingest this data from all those resources? Do not just think of logging and Metrics of your applications and Virtual machines. Also think about your infrastructure network, NAS, etc… But don’t forget your incident systems, your change system has data on incidents and changes in them that is also important for this type of systems. If you have all the data mapped that you want to use the second part is how do you get it to the AIOPS tool. Do they have dedicated plugins, or webhooks, or can they read kafka, or from a database. And for the third part have they automated the data input discovery, or do you need to setup parsing records for all your data inputs?

This type of investigation to the data that you want to use in a AIOPS tool is important because the more data you can correlate the more valuable the tool becomes.

Analytics on real time data on the moment of ingestion.

• Can they handle the amount of data you want to send them and they do real-time analytics on this data? • What will be the maximum delay when you do 100% of your data or 150% of your data?

Historical analysis


Historical analysis of stored data.

• Can they look into the historical data you have and use those to build models? • Can they see trends in the historical data that would help you?

The real-time analyses are important but you have a lot of historical data so they should be able to use those.

Provide access to the data.

These AIOPS systems are going to produce a lot of data them self. So, it stands to reason they will provide a GUI to it. • Check if the GUI they provide is good enough for you? • Is the GUI customizable for your needs or is it one type fits all? • How is the security to the data arranged can the data be security at all?

Storing the acquired data.

• How do they store their data is it in a dedicated database, or is it in an open database? • Do you have the expertise in your company to support a database that they use? • Do you want them in a cluster that you have already implemented in your company and can that cluster handle the load?

You need to choose what works for you not just what works for the company that you buy the AIOPS system from.

Use machine learning to analyse the data.

What system do they use in there AIOPS tool, is it machine learning or rule based? I would always go for the machine learning.

Being able to take action on result of analysis.

The integration is not just important for the incoming data. You also will need to be able to do something with the results of the AIOPS tools. So, what are the options when the AIOPS tool detects something. Can you page people, email people, create incidents or even setup standard actions on the systems like a reboot or a delete of a file. All these things need to be taken in to consideration.

Large set of integration options.

This is something to think about. If at this moment you choose for a vendor of AIOPS tools and that vendor matches your need of integrations perfectly but he is limited. Then later when you want to switch to a different data stream, like from splunk to ELK, and the vendor does not support it you will have to overcome a bigger set of difficulties. So, if you choose a vendor that fits your needs but also has a bigger set of integrations it is easier to change something in your data streams later on.

Historical analysis

What topic is important.

That will differ from who sees this article but my top 5 is :

  1. Large set of integration options.
  2. Use machine learning to analyse the data.
  3. Analytics on real time data on the moment of ingestion.
  4. Being able to take action on result of analysis.
  5. Provide access to the data.

Conclusion

There are a lot of things to consider, I am 100% sure that I have not hit all of them. But I hope I have started you thinking of what questions need to be asked before selecting a AIOPS tool/vendor.

© 2019 Marcel Koert for MeloMar IT BV