Multimetric ML job Question

I have an index with data of throughput of multiples interfaces, those interfaces have different capacities: 1Gbps, 800Mbps, 500Mbps, 50Mbps.

It will be allright if I mix all those interfaces in a single multimetric ML job or is better to separate them based on their capacity: one job for all the 1Gbps interfaces, another for all the 50Mbps interfaces...etc.

I want to detect sudden drops in the throughput of the interfaces.

Can I assume you want to do multiple interfaces and then for each network element/host? How many network elements are there? For example, if there are 1000 of them and each one has four different interfaces, then that would be 4000 total time series to be analyzed. You can indeed do this with a single job with 2 splits (perhaps partition on hostname and by on interface).

Things might get a little unwieldy if the number of total time-series in a single job gets "big" (as in more than 100k series) - one big job is a lot less flexible than distributing multiple, smaller jobs to multiple ML nodes.

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.