how-to block ads
|
|
Share Topic
 |
 |
|
|
Ulmo
join:2005-09-22
San Jose, CA Reviews:
 ·SONIC.NET
 ·callwithus
·Vitelity VOIP
| reply to devnuller
Re: Give Me A Break said by devnuller:usage of 6Mbps running full out for only 5% of the month (95%ile billing that commercial uses today) Could you clarify that? What is 95 percentile billing? Here's what I found in Google so far:
»www.epidirect.com/Bandwidth/Unde···tile.htm
Are those representative?
So, in other words, they ignore 99.999% of your use? That seems dumb. Why not count it? I've created metrics that do stuff like this, but what help is this particular metric (that ignores most of your use)? Is it outdated or overapplied?
Reading their example, I see that:
1. (Top) 5% is ignored. Ignorance is bad.
2. (Bottom) approaching 95% is ignored after that. More ignorance is worse.
Almost 100% of your use is ignored; only one essoteric sliver of your use is considered. That is insane. I would never consent to a charge plan like that!
They ought to do flat-per-use charging, not log scale or anything obtuse like that.
For instance:
* You are charged the sum of the following:
A. $Y per Z of circuit bits per second capable.
E.g., you have a 1 Gigabit Ethernet connection (10^9 bits/second each way full duplex), and you are charged $1 per day for it ($1 per Gigabit per second bandwidth connection per day). That is about $365.24 per year. They could even charge you just a one-time flat fee of $7,283 to install, once. (Or, what about only $823.28?) Different options abound when you do cost-based charging rather than insane subsidizing.
B. $W per actual X bits used in bandwidth out the end of that connection (they could charge different rates for inbound and outbound, but I'll assume they're charged the same for this example).
E.g., you use 100 terabits (of your ~63.1 available petabits, or half that each direction) in a year of usage, and you are charged approximately $10 per terabit, so that is a charge of $1,000 for usage per year, or $83 per month, in addition to the above charge for the connection.
No ignorance; every bit counted.
For comparison, but not that it matters (it could all be bursted in one day I guess), is the average use of the circuit would be ~3,168,896 bits per second (of both directions), so half of that is ~1,584,448 bits per second, which is 0.158% of your gigabit ethernet connection's speed in one direction, or 0.317% of your gigabit ethernet connection's speed in one direction if all of the bits went the same direction.
If you had a 10 megabit/s connection, then that would be about an average of 32% of your connection used, once again, irrelevent, since you could use it all at the same time or not.
If the ISP was charged more for peak times and less for off-peak times, then they could pass those charges to you, but that doesn't make sense; they ought to just charge one rate for every bit at every level. We don't need to subsidize the peak or bursting users when we're just non-peak and non-bursting profiled users. If the peak based or the bursting based users want more bandwidth for their peaks or bursts, then they should pay for it, not us, the constant users. Anyway, the bursters get to benefit from an aggregate of bursting users all paying for things together. The peak users are the ones that might bitch and moan; they are the ones that ought to either shift their use elsewhere (like a service that has sufficient peak availability), use less, or use at a different time. There is no need for us to subsidize them.
The cost per bit can be calculated by the ISP on a daily basis. M bits passed this connection today, and the connection costs O dollars per day in total installation and maintenance costs over its entire life averaged out, so the cost per bit today is $O/M. That is, the more bits that passed, the less each bit costs. Then, a margin level (profit) is applied to that amount (let's say 2%-15%), and then that amount is the charge applied that day. Since that is hard to track by users, there would have to be a buffer: instead of per day, it would be done per month. That is, cost per circuit per month is $N, and the bits used in a month is P bits per month, so $N/P times multiplier (say 15%) is how much bandwidth costs the next month, as an estimate based upon the previous month. The ISP could quote you double according to previous month for your limiting budget metrics, then charge you actual rate for that real month's usage, too; that would be pretty simple. The ISP could eat any undercharges due to overdouble growth rates for one month, and they can temporarily surcharge the rate to adjust, realigning within one month.
That's a lot better than "percentile" charging, which is *TOTALLY* bogus. | |
 espaethDigital PlumberPremium,MVM
join:2001-04-21
Minneapolis, MN
kudos:2
 ·Clear Wireless
| said by Ulmo:Could you clarify that? What is 95 percentile billing? Here's what I found in Google so far: » www.epidirect.com/Bandwidth/Unde···tile.htmAre those representative? So, in other words, they ignore 99.999% of your use? That seems dumb. Why not count it? I've created metrics that do stuff like this, but what help is this particular metric (that ignores most of your use)? Is it outdated or overapplied? Reading their example, I see that: 1. (Top) 5% is ignored. Ignorance is bad. 2. (Bottom) approaching 95% is ignored after that. More ignorance is worse. Almost 100% of your use is ignored; only one essoteric sliver of your use is considered. That is insane. I would never consent to a charge plan like that! The problem is you are thinking of it in terms of quantity (number of megabytes or gigabytes) when 95th percentile measurement is about rate. Your key limitation on the ability to deliver network capacity is the speed of your interfaces, which is again determined by a rate in bits per second.
95th percentile calculation is all about "What is your peak 5-minute average utilization most of the time?" This allows the carrier to determine how much capacity to build out, and how to bill their end-users for the bandwidth.
The calculation is done by taking a sample of the number of bits moved across an interface every 5 minutes. Every 5 minutes the difference between the numbers is divided across the 5 minute interval to get an averaged bits per second rate. This evens out spikes from short duration transfers. At the end of the month the rate samples are lined up from greatest to least, the top 5% are discarded, and the next highest rate number is used for billing. So if your 95% value is 2.2mbps, you would be billed for 2.2mbps of transfer rate in addition to your access charges.
The reason you do this instead of billing by quantity is that from a resource utilization standpoint, someone who transfers 300GB in one day consumes more of a circuit than someone who transfers 300GB over 30 days. To move that quanitity of traffic over 30 days would only require 1 mbps of traffic, but to move it in a day would require nearly 28 mbps.
This is the de facto standard for Internet billing, and EVERY carrier bills this way. ATT, Verizon/MCI, Sprint, Level(3), Savvis, Qwest, TimeWarner Telecom, TeliaSonera, GlobalCrossing, Cogent, and every other carrier I haven't mentioned.
-Eric | | |
|
Ahrenl
join:2004-10-26
North Andover, MA
1 edit | These rate charges are essentially setup to cover maintenance, profit margin, and future build-outs. Greater utilization of the network shouldn't materially increase the maintenance costs for the carrier's, but will substantially increase revenue available to increase the second two buckets. Therefore, in a competitve environment, rates should come down as utilization ticks up, at the same time as greater build-out revenue is generated.
I guess I don't see the problem. | |
Reviews:
 ·Comcast
·Charter
| said by Ahrenl:Greater utilization of the network shouldn't materially increase the maintenance costs for the carrier's, but will substantially increase revenue available to increase the second two buckets. Greater utilization = more capital to add capacity = more maintenence costs for warentee's of that capital
Significantly greater utilization = lighting more fiber = significantly more capital = major router upgrades = more facility space = more power = more maintenence costs | |
Ahrenl
join:2004-10-26
North Andover, MA | said by devnuller:Greater utilization = more capital to add capacity = more maintenence costs for warentee's of that capital More capital will be needed to add capacity, but I'm talking about the long haul resellers. They'll be earning greater revenue before more capital will be needed, thus providing the capital through retained earnings. Maintenance costs should only materially increase if you expanding the size of the network.
said by devnuller:Significantly greater utilization = lighting more fiber = significantly more capital = major router upgrades = more facility space = more power = more maintenence costs There's nothing new here from your last point. Again, we're not talking about last mile operators here. | |
Reviews:
·Comcast
·Charter
| said by Ahrenl:More capital will be needed to add capacity, but I'm talking about the long haul resellers. They'll be earning greater revenue before more capital will be needed, thus providing the capital through retained earnings. Maintenance costs should only materially increase if you expanding the size of the network. It is true that more bandwidth=more revenue=reinvest portions into capacity growth for the first mile and middlemen ISPs (who charge based on peek usage).
I think that is the point of a need to charge based on usage tiers (not bill-by-the-byte). If there is a drastic increase in demand, the revenue will follow to address the capital demands. | |
|
