Of course Telkom is going to tell you to use the portal that shows the higher figures. It clear that they base their FUP levels on the higher numbers.
The portal shows the usage as recorded in the usage management system, that implements the FUP, as that is the *MOST* relevant to a customer.
It also (under connection history) shows the usage reported by RADIUS, and allows you to download spreadsheets of both, so you can compare.
They should correlate very well, but won't be identical.
5 years ago it was decide to stop using RADIUS for usage enforcement,as:
* Telkom is not allowed to disconnect users unless they are over quota, so it is impossible to do accurate accounting of NightSurfer using RADIUS
* The regulatory status of disconnecting users for FUP enforcement was unclear (the regulations were the result of the MyADSL initiatives to try and get regulations approved in ~2004 that would require Telkom to make all DSL uncapped, and failed).
* Openserve doesn't offer CoA (change of authorization, to change the profile of a connected subscriber without terminating their PPP session).
* No other smart features (e.g. the bolt-ons that we have offered) are possible
Afrihost decided to stay with RADIUS, and they disconnect users to change the shaping profile applied by Openserve, and also have inconsistent accounting of their NightSurfer feature.
The Telkom Internet usage management was very reliable for about 4 years after it was launched (in Feb 2013). There were some issues, but most were rectified (and usage corrected or topups given) within a day.
However, the operational support for Telkom Internet has suffered from the restructuring at Telkom, and the retrenchments at BCX.
Also, check that they don't randomly change your reset date to a different date.
It isn't done randomly, it is done once, and clearly communicated to you when you reach FUP3 and accept the FUP message.
Two months ago, I stayed on FUP 3 for an extra week.
You shouldn't have. Your usage should have been reset when your rollover date changed. I wrote tests for this when I implemented it, so it was definitely working (though it could have broken in the past 2 years).