Math formula in trends

[limo]

Hi to all,

I am working on multicriterial network analysis using data from Zabbix. During work, I found that it would be nice to have mathematical formulas in trends. Maybe directly in trends or another table.

Today, we know very little informations about values during trends period. But it would be very easy to add math formula which could aproximate data inside that interval. If it is not possible to approximate data, we could use NULL math formula.

For start, maybe polynomical approximation is enough.

D=ax^2+bx+c , so we need only three float numbers. For future use, maybe it is good to add enum column, which could select function. (polynom, logaritm, ...).

I think that many items history can be approximated using polynom with big success. With little amount of overhead, we have great informations about data inside interval.

Theoreticaly, it is not so complicated to add approximation to housekeeper process. Math approximation can be later used for better focused graphs in history or better prediction of data using some external neural network.

What do you think about this?

[Alexei]

You may start with a trigger function which would predict period of time to reach a certain threshold. It would be useful to be notified one day in advance before free disk space reaches 10%, for example. Linear approximation is fine for a start.

[limo]

Yes, simple trigger and prediction could be enough for now. You are probably right, that most of history will be either lineary approximatable or no approximatable.

I am doing further analysis of this and I should know more on Zabbix conference.

Question is, if this should not be possible to activate some function during history deletion in future. Maybe zabbix plugin or another way. To trigger some procedure before deleting data to be able to process them externaly. This should be done directly in C code, because of speed.

Theoreticaly, anybody can write some plugin to store old history externaly, do math operations or another analysis before it is deleted.

Today, I can solve this by API and cron. I will try to do math operations over history and try to approximate it into my table. It is enough for now.

Thank you