In message <CAMD-=vjshvq+g_g0swwj9vdxsvwdktesd0shoqsnka_qozh...@mail.gmail.com>, Nick Edwards writes: > > Hi, > > We have a customer who has their own cache server, but in the afternoons > before they close up for the day, they commit off-site backups, this > process takes them about 90 mins, anyone trying to use the internet in this > time fails 99.9% of the time due to DNS lookup errors, but if they use an > external DNS server, such as ours, it works - albeit slow but it does get a > response. The local DNS cache server operates fine and instant for their > private LAN, and pinging around their LAN is sub 1ms so the problem exists > when bind tries to go out to get answers for real hostnames. When their > internet link is not fully utilized there is no problems. > > The problem arose again today before the off-site backups when just one PC > got its message from Microsoft to grab the anniversary update, at 11 > o'clock in the morning, strangely it did not fill their link, but the pps > must have been rampant because the DNS errors again failed when using their > local cache resolver server. > > Is there a named.conf setting we can suggest they use on their cache server > that perseveres and waits a little longer for answers to send to their > client machines? > They are using bind 9.10.4-p2 with default settings from source package > along with options of - > > directory "/opt/named"; > allow-query { xxxxx; }; > allow-query-cache { xxxxx; }; > allow-transfer { xxxxxx; }; > > > Thanks for any advice. > Nik
There is one word for this. Bufferbloat. This is where the a router has massive buffers for the link and rather than dropping packets when it cannot send packet thereby throttling TCP straight away it queues up traffic creating a very long delay path and eventually throttles TCP to the link speed when the buffer finally fills. I've seen this create multi-second delays in the path. Really bad buffer bloat can create delays that are minutes long. Go talk to your router vendor. This is either a bug in their product or a bug in a upstream router. It is possible to examine the traffic flows in a router and mitigate bufferbloat in another router by resticting the traffic through the first route to slightly less than what the second router will allow. Mark -- Mark Andrews, ISC 1 Seymour St., Dundas Valley, NSW 2117, Australia PHONE: +61 2 9871 4742 INTERNET: ma...@isc.org _______________________________________________ Please visit https://lists.isc.org/mailman/listinfo/bind-users to unsubscribe from this list bind-users mailing list bind-users@lists.isc.org https://lists.isc.org/mailman/listinfo/bind-users