Degradation of service when authentication fails with Windows AD

[Antonio Alberola]

Dear All,
 
I'm having random authentication failures and I think they are due to a
Radius server internal failure. I use Radius for authenticating the email of
users in Windows Active Directory via PAM. Before I used NTLM and Kerberos
together, and now I use PAM.
I use FreeRADIUS version 2.1.12 that includes CentOS 6 repositories. This is
the description of the machine, host x86_64-redhat-linux-gnu
2.6.32-279.14.1.el6.x86_64 CentOS release 6.3 (Final)
 
The problem is as follows. Users authenticate properly during the whole day,
but suddenly authentication begins to fail and user authentication error 
appears even if the credentials are right. Since the failure, the service is
exponentially degrade and it only validates 1 of every 20 requests. The
onset of failure seems to coincide with one of these three messages:
 
Tue Jan 30 08:27:38 2013 : Error: Received conflicting packet from client
localhost port 14038 - ID: 194 due to unfinished request 161451.  Giving up
on old request.
Tue Jan 30 08:27:52 2013 : Error: Request 161507 has been waiting in the
processing queue for 11 seconds.  Check that all databases are running
properly!
Fri Feb  1 14:55:15 2013 : Info: WARNING: Child is hung for request 3609 in
component <core> module <queue>.
 
The solution we are applying at the moment is restarting Radius. Sometimes
restarting does not fix the problem and we have to set Radius for allowing
all connections. Few minutes later, we turn it back to the current
configuration and it works again. The biggest drawback, besides annoyance of
users, is Windows AD accounts are blocked because of the failures.
 
I need help to find the cause of the problem and fix it. I do not know yet
if the problem is in the domain controllers, in the PAM module or in Radius.
But everything seems to point to Radius.
 
Thank you very much.
 
Antonio Alberola.