SERVFAIL under pretty high load and recursive queries?

[Tom Mueller]

Hi,

I have 2 linux boxes running both bind-9.4.3b2 as caching-only nameserver.
The one has a query rate roundabout 1000 queries/sec the other
roundabout 8000 queries/sec and up to and above 2.000 UDP-Ports open at
a time.

I have a third linux box running bind-9.4.2 as authority server only.
On this auth-server I have configured a zone with a wildcard entry.

When I perform a single dig with a name asked never before, the
caching-only nameserver has to perform a recursive query to the auth-server.
This works fine.

When I put this dig command into a for-loop and calculate for every loop
a new name, I get several SERVFAIL's asking the high-load machine.
When I repeat the same for-loop 3 or 4 times, then I get at the end no
SERVFAIL's and bind-9.4.3b2 performs fine.

Same procedure on the low-load machine shows no SERVFAIL's.
Same procedure on the high-load machine under bind-9.4.2 shows no
SERVFAIL's, too.

The high-load machine is running under a 2.6.16.13 SMP kernel and
bind-9.4.3b2 is compiled whith threads enabled.

Can someone give me some hints how to fix this?
Thanks in advance.

Best regards

-- 
Tom Mueller
Entwicklung Dedicated Servers
------------------------------------------------------------
Telefon: + 49 (0)30 - 398 02 - 0
Telefax: + 49 (0)30 - 398 02 - 222
E-Mail:  t.mueller at strato-rz.de
------------------------------------------------------------
Strato Rechenzentrum AG
Pascalstr. 10
10587 Berlin
----------------------------- -------------------------------
Aufsichtsrat: Damian Schmidt (Vorsitz), Stephan Esch, Andreas Sand
Vorstand: Julien Ardisson, Christian Mueller, Christoph Steffens, 
Rene Wienholtz
Amtsgericht Berlin-Charlottenburg HRB 75629