Weird performance issue.
Cedric Lejeune
cedric.lejeune at arcelormittal.com
Wed Aug 20 14:08:40 UTC 2008
Hello list,
We currently running two instances of bind9, each one on a different
host. Both hosts have their own IP address and basic tests work perfectly:
- ping of external server(s) work fine (FQDN and IP address)
- host resolution works fine
- named processes number is quite low (~16)
The problem occurs when we try to move IP address from master server to
slave server:
- ping of external server(s) failed (FQDN and IP address)
- host resolution take a huge time to complete or do not complete at all
(timeout)
- processes number increases significantly (~1000, which seems to
correspond to recursive-clients default value)
We have taken care of everything we can think of:
- bind9 configuration
- network configuration
- arp resolution
- firewall configuration (although being a CheckPoint firewall, Smart
Defense does not seem to cause any issue since only logging is
activated, cf
http://groups.google.com/group/comp.protocols.dns.bind/browse_thread/thread/cfa8c63ec6bd08d6
. Firewall log does not show anything weird too.)
Log do not show anything relevant to me, except the well known "too many
timeouts resolving 'ns2.highergroundtech.com/AAAA' (in
'highergroundtech.com'?): disabling EDNS" message.
We currently running BIND9 on Linux Debian:
- the one running perfectly is a quite outdated 9.2.1-2.woody.1 package
- the one causing problem is a quite up to date 1:9.5.0.dfsg.P1-2 package
Configuration files have only been updated to reflect releases changes.
Do you have any hint or advice so I can at least look at where the issue
comes from and then try to solve it?
Thanks for your help,
Kind regards,
cedric.
More information about the bind-users
mailing list