Hacker News Reader

Comment by Bo102010 | Hacker News Reader

Bo102010

I have worked for ISPs where people who are making decisions about "bandwidth" _do not understand_ the difference between "bits" and "bits per second."

These decision-makers would say that "Our costs for providing backhaul / transport / whatever is going up - no matter how much we upgrade, its capacity is exhausted either immediately or very quickly. We have to recover that cost somehow, or we won't be able to continue providing any service."

However, they don't understand that their equipment and transport are constrained at the level of "bits per second at peak" and not "sum of bytes transferred during a billing period." So, they end up making bad decisions about "charging $x per GB after y GB transferred per month."

View on HN

The reason why we often "charge $x per GB after y GB transferred per month." is that this is simple and the customer can understand it. Usually the idea is that metering costs more than the bandwidth, up to a point, so you get a 'don't worry about it' quota, and then a punishing price after that. Me, I limit people after the 'don't worry about it' price, because, well, the sort of people who are my customer don't want to get hit by unexpected bills.

what would that look like if you had the 'don't worry about it' quota with a more reasonable price on overages? I don't know. It used to be that you'd set the 'don't worry about it' quota a little too high, 'cause with the unreasonable cost of overages, most people would error on the side of not using it all.

(metering costs more for both the provider and the customer, but mostly the customer; the provider side can be automated, so it's largely a fixed cost. The customer usually prefers a flat rate over a variable rate, even if the variable rate was slightly cheaper on average, because variable rate billing means that the customer needs to pay attention, which is expensive.)

When dealing with more savvy customers and selling amounts of bandwidth that matter, charging on the 95th percentile is pretty common.

That said, any major ISP who claims it costs them more than a few pennies per gigabyte to move traffic is flat out lying.

"That said, any major ISP who claims it costs them more than a few pennies per gigabyte to move traffic is flat out lying"

Not lying, but confused (see my comment below) and measuring at the wrong margin.

unless their load is /way/ more peaky than average, they are most likely lying. The people running ISPs know this stuff. I mean, unless their margin is so high that bandwidth costs don't matter, getting different customers with different peaks is what ISPs /do/ (well, it's what transit providers do, and any ISP of sufficient size eventually starts acting like a transit provider.)

I'm saying the engineers may know it, but their bosses _do not_.

It's possible? but I'd be very surprised if this was common. First, it's fairly core to the business of being an ISP; second, it's a fairly easy concept to grasp. I could pretty easily explain the concept to someone who is average; I mean, most upper management looks kinda dumb compared to Engineering, but they are usually above average, even if they usually spend more points on social stuff that doesn't correlate with G.

This would be like saying the president of ford motor company doesn't understand the difference between the EPA 'city' and 'highway' mileage numbers, and how to game each. Sure, if it's convenient, he might claim ignorance to the press, but there's no way he doesn't understand something that central to the business.

How long until we get to peak and off-peak rates for data usage by ISPs, much like some electrical utilities?

That's certainly how it works for most Australian ISPs. And the off-peak times only run from about 02:00 to 07:00. I'm lucky enough to have an unlimited ADSL account with TPG, or my quotes would be a couple hundred GBs for the same cost.

Luckily, I think competition is going to be making things better for the Australian ISP consumer. That, and recent improvements/additions in undersea cables.

I can't speak for your experience with TPG, but a lot of users I know on TPG hate the fact that they have "unlimited downloads" but in reality get limited bandwidth due to an oversold service.

I decided I prefer a limited plan where I know what service level (particularly latency and throughput) I can expect. For instance, plans with Adam and Internode are not that much more than a TPG plan for more usage than I'll forseeably use (I get 500gb a month or so at 20mbps + VOIP for $100), but the difference in gaming latency and actual speeds are both considerably more stable and faster than friends on TPG.

My experience with TPG has generally been very good. I connect at 17-18Mb/1Mb and consistently download at 1.0+ MB/sec. Usually 1.4MB/sec, though I've seen it drop to ~800KB/sec.

Given that your plan is 60% more than mine, I'll gladly pocket the difference and live with the minor problems I've had. As another poster mentioned, I chose to vote with my wallet. Generally small quotas + absurd off-peak times (where the majority of the quota is allocated) make my blood boil and I'll support anyone who helps break up the status quo.

Off-peak times vary betweem ISPs. But you can also vote with your wallet. I'm with Internode, which is a little more expensive, but they don't have peak/off-peak (among other advantages, for me at least)

It doesn't translate to direct costs, but my speed is largely dependent on contention with other users, and drops heavily during 'peak hours'.

Then again, I suppose you could argue cost per GB = integral(download speed * time) / cost per month, so 1GB at slow speeds is costing me more in terms of time-bandwidth product at peak rates than otherwise. Soft data caps and peak-time QoS might also be happening behind the scenes, skewing matters more.

Cost per GB isn't really even what you want to measure, and doing so is what causes so much confusion.

Let's say you're an ISP and you have a 10Gbps connection to the outside world, and the routers etc. that can keep up with that speed.

To a first approximation, your costs to serve up 10Gbps to all the customers in your area is 0 (you've got to pay for electricity and engineers and etc, but this doesn't vary with user activity).

If your customers in your area all queued up nicely and downloaded stuff at full rate, never overlapping with each other, you could transfer 10Gbps * 1 month for basically nothing.

However, your customers all want to do stuff at the same time, and thus contend with each other for a slice of that 10Gbps. If you want to serve all of their requests, you're going to have to spring to upgrade your total capacity.

If you don't own your own fiber, you have to rent a link from someone else, or if you do and there's no more available you have to physically come up with more (and pay to dig up the road and etc.). Now your routers and etc. might not be able to keep up with your new backhaul speed, and you have to upgrade those...

You really want to measure "Marginal cost of +x Kbps during the busiest typical hour."

Probably quite a while. Phone companies have done and continue to do this, but are moving more and more away from it.

I'd be thrilled if someone tried it, though. It's more honest and more likely to address bad user experiences than arbitrary "bytes transferred per month caps."

I have this now. It seems to work pretty well to encourage heavy use of the connection out of hours.

That's how http://aaisp.net is billing (but they're quite nerdy in general, e.g. you get your own IPv4 with personal RIPE entry and RevDNS + IPv6 ranges)

And some of them don't even understand basic arithmetic: http://verizonmath.blogspot.com/2006/12/verizon-doesnt-know-...

Reply on news.ycombinator.com