There’s a known issue with a particular ethernet driver (e1000, I think)) in Linux. We pull the numbers for the chart from /proc, and there’s a bug with that driver and randomly multiplying reported bandwidth. It’s unfortunately a bug in the driver/kernel.
The charts are definitely not the most accurate. We pull snapshots of bandwidth, CPU, etc, every 5 minutes via CRON, and no attempt is made to determine what may have happened in the other 4 minutes and 59 seconds. It’s primarily a “quick glance” thing, and we want as little impact on server performance as possible to get it.