Sure - The route cache was largely removed in (IIRC) 3.8, but there's still entries that get stored[0]. There's a limit to how many entries Linux will store, a

Sure - The route cache was largely removed in (IIRC) 3.8, but there's still entries that get stored[0]. There's a limit to how many entries Linux will store, and like any LRU-esk data structure rapidly cycling entries through it isn't going to do anything wonderful for your performance, never mind if you actually expected to use any of the cached data for a business performance 'feature'. 25g NIC is an awful lot of 60byte packets. I'm not saying this is going to be a common concern, just that like any other shared kernel resource cgroups and namespaces aren't going to help. 0: https://www.systutorials.com/docs/linux/man/8-ip-tcp_metrics/ https://www.systutorials.com/docs/linux/man/8-ip-tcp_metrics...