search

vacp2p / wakurtosis

4 stars 3 forks source link

Alberto/cadvisor bandwidth into master #130

Closed AlbertoSoutullo closed 1 year ago

AlbertoSoutullo commented 1 year ago

Quick change to cAdvisor plotting.

Before this change, "by_simulation" in config.json ment that these metrics would be aggregated per timestamp.

For example, if we have 60 timestamps and 10 containers, for each timestamp, I was aggregating the metric for all nodes in timestamp 1, then in timestamp 2... etc.

This was my bad since I understood this wrongly when we talked about "accumulated" values.

Now this is fixed. Values in config.json "by_node", will behave correctly. Bandwidth and Disk I/O are already accumulated in cadvisor, so we are just getting the maxmum value from all timestamps, and this is the plotted value.

Remember that cAdvisor does not divide metrics per container.

0xFugue commented 1 year ago

Am glad it is resolved. Just a qq: how do you mean "Remember that cAdvisor does not divide metrics per container."? AFAIK, at every epoch/timestamp, cadvisor collects per-containers stats from all visible containers, right? May be I misunderstood 'divide'.