'cacti' isn't really a monolithic thing. Ultimately it's a gui front end for rra files and rrdtool. If one chooses not to go down the route of disk space intensive but lossless time series database interface metric storage (influxdb or similar), we are talking about what level of detail is lost over the week and month scales in an rra file.
The step settings and compression-over-time for the rra file, defined when its first created, will affect precision as well as the poller interval. An older default cacti install with 5 minute intervals and small rra files (under 500KB max size for the whole time scale per single interface and file) will look very different than on a cacti 1.2+ install set up for 1 minute poller intervals. If you do a new cacti install on Debian testing or unstable, during the initial configuration process you'll have the options to define these. The difference between the two is best seen on some sort of circuit that sees lots of very brief bursty high Mbps/Gbps traffic.
In 2020 considering the low cost of disk space I would recommend something time series database storage based, polled on 60s intervals for interface bps. Even if each discrete 95th billed interface eats up several hundred MB of disk space. You can always set up something later on, to truncate the time series db to throw away all data older than 12 months.
hi there,
i have asked about this in the past. What is the best tool out there to do 95th percentile billing. I have decided to use observium and librenms as result of responses but there seems to be some kind of billing module issue with these tools (thy are basically the same code).
What are other systems besides observium and librenms (and old fashion cacti) people are using these days with 95th billing and integration with a CRM like salesforce/zoho, etc. I appreciate the responses.
Mehmet