[Serusers] FW: radius accounting problem.

[Greger V. Teigre]

Dear Lucas,
Basic RADIUS accounting uses start and stop.  RADIUS is not one protocol, 
but rather an umbrella with a collection of extensions.  Many standard 
attributes are not used an most applications and many non-standard 
attributes (vendor-specific) are used in many applications...
    Most service providers prefer to use start and stop records and use a 
billing system to calculate the session time.  This is because it allows 
pre-screening of faulty/problem sessions, standardization/consolidation of 
records across accounting sources and one can better calculate rates, i.e a 
phone conversation starting 7.35am and ending 8.10am may be rated with one 
rate from 7.35am to 8.00am and another the remaining time.
    Yes, some accounting sources send Acct-Session-Time, which for some 
applications is useful because you don't have to analyze and match start and 
stop records.  However, if you don't have the attribute, it would not make 
sense to modify your RADIUS server (I don't know IC-Radius, but in general) 
because the RADIUS servers purpose is to acquire the records, while 
manipulating the records is a completely different problem domain that is 
better left to billing systems.  Of course, some RADIUS servers may come 
with simple tools to match start and stop and produce raw CDRs (Call Data 
Records), but this is a tool that is/should be used batch-wise and not 
real-time, IMHO.

    When it comes to ser, here is my understanding (correct me somebody if 
I'm wrong):  Even though the tm module implements stateful transactions, 
this does not mean it is keeping the states of on-going calls, but rather 
that each SIP message is treated statefully. A simple example is that a 
resend of a message can be recognized.  An accepted INVITE will in 
accounting terms generate a start, but ser will not keep track of on-going 
calls in memory, so when a BYE arrives, a stop will be logged, but ser has 
no information about the matching start of the call.  Of course, ser *could* 
lookup the matching start record in the database and calculate the 
session-time. However, this would seriously harm the scalability and is 
against the design of the system.  I believe Asterisk keeps the call state 
and of course IP-PSTN gateways with open PRIs during a session will have to 
keep the state and will thus be able to calculate the session-time without 
performance penalties.

    I hope this helps.
g-)

Lucas Aimaretto wrote:
>>> Your radius server can keep track of the call duration
>>> by subtracting the stop time from start time, you do
>>> not need per se the header you mention nor a B2BUA.
>>>
>>> See www.freeradius.org documentation or the sample
>> configuration file
>>> to achieve what you want at http://cdrtool.ag-projects.com/sql.conf
>>>
>>> Adrian
>>
>> The thing is that I know many applications, like Cisco 5350,
>> SNOM SIP SERVER ( www.snom.org ), that send the
>> Acct-Session-Time to RADIUS. In fact, RADIUS (at least mine,
>> IC-Radius) is ready for receiving such attribute. Why should
>> not SER send this standar attribute?. If I want my RADIUS to
>> estimate the Acct-Session-Time, I need to touch my RADIUS'
>> code. And touching code means not a standar way of achieving
>> accounting, because, I can modify my code, because it is open
>> source (IC-Radius). But what would people do if they do not
>> use IC-Radius or FreeRadius? I do not want to modify my code
>> ( unless there is no other way of getting the session time).
>>
>> Regards,
>>
>> Lucas
>
> Am I wrong in my thoughts ?
>
> Regards,
>
> Lucas