[bitfolk] Additional bandwidth graphs being created for some customers

Hi, We were recently made aware that our Cacti¹ bandwidth graphs for a particular customer were dramatically different from reality. On investigation I realised that it was a bug in the Linux kernel where it wasn't using 64-bit counters for Xen backend network devices. As a result for readings above ~228Mbit/s the counter was wrapping twice and reporting incorrect values on an SNMP read (as used by Cacti). This is a fairly minor issue because we do not use SNMP counters for billing. It does mean that if your VPS has ever done more than about 228Mbit/s average in a 5 minute period that Cacti won't be showing it properly. The kernel bug has since been fixed but deploying a fixed version would involve using a self-compiled² backports kernel. I am not going to do this because I haven't tested it enough yet. Instead I have identified the 30 or so customers that have ever recorded that much bandwidth use in the last 12 months and am adding new bandwidth graphs for you, using 1-minute polling. Also new customers will have the 1-minute resolution graphs. That should be safe to about 1.1Gbit/s. So, if you are looking at Cacti and see you now have two bandwidth graphs with one cutting off where the other began, this is the reason why. I wrote a blog article about this at: http://strugglers.net/~andy/blog/2017/09/03/when-is-a-64-bit-counter-not-a-… Cheers, Andy ¹ https://tools.bitfolk.com/cacti/ - Log in with your usual BitFolk credentials ² We already use self-compiled kernels based on Debian kernel packages, because some security patches have not yet made it into Debian's packages. Building the kernel isn't the problem, it's testing it well enough. -- https://bitfolk.com/ -- No-nonsense VPS hosting _______________________________________________ announce mailing list announce(a)lists.bitfolk.com https://lists.bitfolk.com/mailman/listinfo/announce