Can anyone shed light on how ISPs handle missing samples when calculating p95s for monthly billing cycles? Do they fill null samples with zeros or leave them as null?
I’m working on a billing sanity tool and want to make sure to cover my corner cases well.
Thanks!
Dan
Interesting question. How often are you missing data? I’d expect that to be pretty robust.
I can say that missing samples weren’t back filled when we billed. Never had any complaints.
-Steve
From my provider days if you miss data you can't bill it or assume zero.
Mike
You have to zero it.
-Ben
As per our village lawyer and accountant …
Assuming
95th percentile billing, sampling every 5 minutes.
You’ll need about 1.5days worth of 0 (~447 samples missing in a row) to bork the curve… and it goes the bith ways.
If you’re 5Gbps committed on a 10Gbps and you burst 10Gbs for 1.6 days, you pay for 10Gbps for that month.
And yes that include the best effort to manage DDoS from both part. We’re capitalist, but no one is from big pharma =D.
Background: I used to own the code that was used to bill for awhile...
You can calculate the min and max by running it twice. Once with zero for missing values and once with max line speed.
The true bill will be somewhere between min and max.
If you are the ISP you would need to assume zero. You can only bill what you can prove.
If you are the customer, you need to assume max. You can only claim excessive billing for what goes above what you can prove.
Regards
Baldur
tor. 28. feb. 2019 04.32 skrev Daniel Rohan <drohan@gmail.com>: