Re: Concurency Revisited

Michael Whisenant ( )
Fri, 08 May 1998 08:53:55 -0500


Actually I am running 4 RadiusNT servers. One the 2 that are
authenticating I have them alternating between NAS, and the two that are
running accounting are alternating as well. I went to this method a few
months ago thinking that was a part of the problem and per Dale's
recommendation. The database is getting less than 30,000 call records per
day, yet I have seen on a few instances inaccuracies in the calls on-line
table. Very few, yet more than I feel comfortable with.

At 09:28 AM 5/7/98 -0400, you wrote:
>I've been running 2.5 for about a week now using concurrency control, and
>it seems to be working real well. What could be happening (I'm assuming
>that you are using this with your USR boxes Mike) is that you are using
>only one radius server. I've found that using multiple (right now I'm
>using two) and staggering which one is primary on my different USR chassis
>and that cleared up the problems with errant callsonline readings. From
>what I've noticed is that RadiusNT will drop calls when it gets too many at
>a time (and can cause end sessions to be a bit better).
>Hope this helps!
>At 05:24 PM 5/6/98 -0500, you wrote:
>> I have a new wrinkle. I log all call failures to RADLogs and have just
>>started using concurency control. It logs the error as over the limit and
>>that x users were already logged on to the system with that IP address.
>>What would be really nice is that it logs the IP address, or the
>>NASIdentifier, or the NASPort so that one could confirm the situation.
>> Reason.... I have had three users that were being failed as over the
>>limit. I happened to be watching, and went to calls online, figured where
>>the user was reported to be online, telnetted into the box and guess what.
>>The port was empty. Now three confirmed mistakes in 18 hours is a bit too
>>much for my blood. Has anyone else had problems with the concurency
>>control and the accuracy of the information? Now the SQL database is
>>running fast, the average time to update an accounting packet is less than
>>200ms so how can calls table not be accurate? Does this also mean that the
>>records that I use to bill user inaccurate?
>>Michael J. Whisenant
>>Vice-President, Operations
>>AIRnet Internet Services, Inc.
>>ph: (256) 704-4692 fax: (256) 704-2329
>> ----------------------------------------------------------
>> RadiusNT Mailing List
> ----------------------------------------------------------
> RadiusNT Mailing List
Michael J. Whisenant
Vice-President, Operations
AIRnet Internet Services, Inc.
ph: (256) 704-4692 fax: (256) 704-2329