I hope you don't mind my 'hijacking' your question and putting it up
on the RadiusNT list, but it is asked often enough and in enough
ways that I believed it bore repeating here on the list.
I believe fundamentally, the problem is that by design
Radius accounting records are UDP packets. UDP stands for
Unreliable Datagram Packet. The accounting stop record may
never make it to the Radius server and hence the database. We've
done everything we can to optimize on the accounting end and
you've done the right moves below in adding to your SQL Server's
muscle. However, there's nothing that can totally stop that, "only a
few percent of the time" problem from occurring, especially when
the UDP packet is in the bit bucket. That's why the Enterprise
version of RadiusNT has SNMP support. Unfortunately not all NAS
devices have the SNMP MIB support necessary to have the Radius
Server query for port by port access. If you have one of the few
which supports this functionality, great! If not, beat on your vendor
a bit and tell them you want it in an upgrade. Mention us. Mention
their competitors who do. :_)
On 15 Dec 98, at 12:17, Greg Hastings wrote concerning:
Users online problem
> We are experiencing a problem which has been occurring now for sometime.
>
> A few percent of the connected users are not showing up online in Emerald
> (version 2.1.11). And, a few percent of the users who have disconnected
> remain showing online.
>
> I believe we have the latest calls_insert trigger installed on the calls
> table.
>
> The problem appears to be consistent, but random (i.e... doesn't seem to be
> associated with any particular ports). At any time of day, I can check the
> users online and compare it to the actual connections. Typically there will
> be from 16 to 32 users connected. Almost always there will be 1 or 2
> connected users who don't show up, or the other way around.
>
> We are running RadiusNT version 2.5, which we recently upgraded to. The
> problem was occurring steadily on the previous version we had been running
> for about 1 1/2 years.
>
> I recently increased the memory on the RadiusNT server machine to 98mb. I
> increased the SQL Server 6.5 allocations to 48mb for memory and 32mb for
> tempdb ram. This is on a 133mhz Pentium, and the only thing running on the
> machine is SQL Server and RadiusNT (as a service). That didn't help.
>
> I have tried running RadiusNT with the -x15 switch from the command prompt.
> But didn't learn from that. The errors occur at random, and are only a few
> percent of the time. So it is hard to spot a particular problem that way.
>
> Any suggestions on how we can proceed in diagnosing and solving this
> problem. Because of this problem, we can't limit our subscribers to 1
> simultaneous logon. And our usage information (primarily the TSGraph) is
> distorted because of the incorrect accounting.
>
> Regards,
> Greg Hastings
>
>
Jeff Presley - IEA Software www.iea-software.com
- Billing and Internet solutions today --
509/444 BILL ext 13, 509/624-9903 (fax)
For more information about this list, including removal, please
see this URL: http://www.iea-software.com/maillist.html