My company builds bespoke PC based routers. Currently these are all based on Centos 7 however I am attempting to port this to CentOS Stream 8.<br />
<br />
Typically I am using RP-PPPOE, widedhcpv6 bind dhcpd iptables tc etc.<br />
<br />
Topology is typically an internet connection -> PC based router -> VLAN tag interfaces -> switches/access points -> clients.<br />
<br />
Nics are mostly Intel NICs using igb driver.<br />
<br />
On my test network I've been investigating unusual performance degradation impacting Android and iOS devices. I've found that IPv6 connectivity is intermittently failing for impacted wired/wireless devices; this is presumably causing the weird delays I perceive when using my mobile phone (yes I also tested with hard wired android devices). I've not been able to replicate the fault using wired/wireless Windows/Linux desktop devices.<br />
<br />
During my testing I would wait for an android or iOS device to get into the broken state, I would then start a pcap on various network devices and immediately find that the issue had gone away. I spent *many* hours checking firewall rules, updating software, firmware, resetting devices, various settings all manner of different things. Eventually I reverted all my tweaks and placed the interface into promiscuous mode and found that the fault did not recur.<br />
<br />
I spent ages trying to think about any filtering 'above' the firewall which might be disabled in promisc mode and cant think of anything. Therefore I have settled on this likely being a kernel bug or something 'weird'. Any input would be appreciated?
↧
