Slashdot Mirror
ISPs Fight Against Encrypted BitTorrent Downloads
oglsmm writes to mention an Ars Technica article about a new product intended to detect and throttle encrypted BitTorrent traffic. When torrents first saw common use ISPs would throttle the bandwidth available to them, in order to ensure connectivity for everyone. Some clients began encrypting their data to get around this, and the company Allot Communications is now claiming their NetEnforcer product will return the advantage to the ISPs. From the article: "Certainly, increasing BitTorrent traffic is a concern for ISPs. In early 2004, torrents accounted for 35 percent of all traffic on the Internet. By the end of that year, this figure had almost doubled, and some estimate that in certain markets, such as Asia, torrent traffic uses as much as 80 percent of all bandwidth. However, BitTorrent is an extremely important tool that has many uses other than what everyone assumes it is good for, namely movie piracy."
many uses other than what everyone assumes it is good for, namely movie piracy.
- Game Demos
- Software updates / upgrades
- Free / Legal Videos
Look, I use Bittorrent and it's great. But I also run an ISP.
The thing is, bandwidth isn't cheap. People bitch that ISPs "oversubscribe", and that we can't really deliver our advertised bandwidth to everyone all of the time. This is true, but how do you think we manage to sell people 5Mb connections for $40/month? Do you know how much 5Mb of bandwidth costs and ISP? It's a lot more than $40. In the market I'm in, we pay THOUSANDS of dollars for that much bandwidth.
The real problem is that bandwidth is too expensive in this country, thanks to the likes of AT&T and MCI and all the other big players. They've got tons of unused fiber lying around, and it costs them next-to-nothing to use it, but it still costs the end-user (in this case, the ISP) a hell of a lot of cash.
Exactly.
The price is formulated on the basis that you do not use it.
I agree with you - this is fraud and there is only one way to fix this.
The problem will go away immediately if ISPs turn off flat pricing and users start to pay for bandwidth used. Even better - if they start charging a differential/tiered pricing depending on the type of traffic. There is no rocket science here. The gear currently on the market is supposed to be able to do it (does it do it is a different matter).
The business models is well known and this is the way the Internet used to operate all the way up to the end of the 1990-es (especially in the slower peripheral parts). This was abandoned when the incumbent telcos entered the access market in the end of the 1990-es. They went after scale and port densities which resulted in bandwidth accounting features being abandoned across most of the equipment. Cisco broke all of its accounting by introducing CEF, other vendors were not any different.
Over the last 5-6 years most of the features crept back due to demand by business users so technologically the gear is in the same (or better) shape as before the telcos entered the market as far as accounting is concerned. In addition to that new gear from Ellacoya, P-cube and such can do things the old systems were not capable of.
All it will take to get this working now will be people who know how to formulate a viable product and tie this up all the way into billing, CRM and relevant backend systems. Unfortunately there are not that many people left capable of doing it in most ISPs so they prefer the BIG STICK(tm) or the "magic vendor silver bullet". It is easier. It does not require investment. It does not require thinking. It does not require competence. Sad, but true - this reflects the state of the industry.
It is rotten, it sucks and it hates its customers.
Baker's Law: Misery no longer loves company. Nowadays it insists on it
http://www.sigsegv.cx/
Bandwidth accounting isn't necessary.
I work for an ISP. Yes, we oversubscribe. It's the way the business works. We only see problems when many people use their bandwidth *at the same time*.
Moving more data total does not cost any more many than for the electricity to move it. What costs more money is having more available bandwidth so that more can be moved at one time.
We get our bandwith from first-tier providers. They do not charge us by the amount we transfer, but they charge us for the speed of the port. They don't care how much we transfer in total, they only care how much they use at once. We do likewise for our customers, with the exception that we oversubscribe.
Oversubscribing doesn't cause problems as long as there's enough available bandwidth out and the hardware to handle it. Some people expect dedicated bandwidth, and for them there are the options of lower speeds or more money.
I want to see oversubscription come to an end, but I don't see it happening. The dropping price of bandwidth and network equipment is primarily driven by increasing customer demand for higher speeds rather than by an increased number of customers. Unless prices drop as customer demand for higher speed remains static (or at least grows slower than the prices drop), dedicated bandwidth at today's consumer-appropriate speeds and prices isn't going to happen.
I've heard this before, and I'm not sure I buy it. Let's say 3 Mb/s costs $60/month. I see that Cisco's 12000 series router go from 2.5 Gbps to 10 Gbps. Assuming that Cisco is being honest about their bandwidth capabilities (e.g. not lying through their teeth like a broadband service provider), that means that a single low-end Cisco 12000 series router can service about 800 customers (assuming that each one actually saturates the pipe 24 hours a day, 7 days a week), each paying $60/month, which equates to $50,000/month in revenue. Now, Cisco doesn't tell you how much these things cost (or even hint at how much), but lets say one router costs a (ridiculous) million dollars. In well under two years, the provider will have recouped the cost of the router itself. Even if the router lasts only a measly year after that, the provider clears an additional $800,000 on their initial investment to cover paying the admin staff (over three years, probably $600,000), power bills, rent, etc. That's pretty close to break even, if the router cost $1,000,000 and only lasted three years (somebody around here has to know what they cost and how long they last - I'll bet it's a rosier picture than I've painted). So I figure $60/month must cover the actual costs they'd incur if we all used the bandwidth we pay for (which would be almost impossible, even for a die-hard torrent user) - I find it impossible to beleive that they'd need to charge $600/month to turn a profit.
Proud neuron in the Slashdot hivemind since 2002.
Perhaps a better analogy can be found in the airline industry (also a service). Historically airlines have routinely oversold seating because more often than not it works out for them. Some people will cancel the fight, some people won't show up for the flight, and sometimes they won't be able to sell all the seats in first class and can bump overbooked coach passengers to first class. In the event that they can't put you on your purchased flight, they will put you on the next one, or refund your ticket. Either way generally sucks for you, but you're at their mercy. So there is at least one industry that has been overselling a service for a very long time.
Slackware, what else when it must be secure, stable, and easy?
> lets say one router costs a (ridiculous) million dollars
It's not that ridiculous. In fact, I'd say you're low-balling the cost by quite a bit. And if you want to have redundancy (no one likes having their service disrupted for days while you're waiting for a replacement card), you can start doubling that automatically. Not only that, but you're not accounting for the cost of doing anything with those connections. A local ISP has to buy service from one or more of the Tier 1/2s. Oddly enough, purchasing an OC-192 (that's that 10 Gbps pipe) isn't exactly cheap. Considering most of the world's backbones consist of OC-48s and OC-192s, and considering that the backbone providers don't want to oversaturate their own lines, they charge the local guys a heck of a lot for that OC-192. No local ISP could ever afford to purchase an OC-192 just for 800 users, and no backbone provider could ever support it as well.
The pricing worked rather well when people were only downloading relatively small files periodically. As long as traffic is bursty, that is. It's when people start downloading large files (like movies) constantly where everything goes awry. If you honestly expect to use that cable providers 5 Mbps down, 1 Mbps up service at $60/mo, when they in turn have to purchase 4 T1 circuits at ~$500/mo to support you, you deserve the crappy service you get. If you want to push that much traffic constantly, buy the T1s yourself.
think traditional telephone companies.
Okay...
They also provide, for a fixed monthly fee, unlimited access to the telephone network.
Hardly. They offer very limited access to the telephone network--you can make and receive phone calls with a limited finite set of optional features such as caller-id and voicemail. They offer unlimited use of that application within, well, limits, including geographical toll boundries and pay-per-use products such as directory assistance and three-way calling.
IP networks offer an ever-expanding variety of access, limited only by the contractual terms of service that each customer agrees to at the time of purchase. In practice, those terms are most often loosely enforced, if at all, and usually only in response to some operational problem caused by a violation. New network applications are developed and widely adopted as time passes.
If they operated on the same principle as the ISPs, you would get nothing but busy signals if more then 0.1% of people decided to call each other.
Actually, telephony capacity is engineered to some threshold of dropped calls per 100 at the network's "busy hour". This threshold is either dictated by regulatory bodies or is left to the telco. Either way, few--if any--telcos build to "zero drops per 100 at busy hour".
Telephony networks are a smidge easier to engineer from a capacity perspective because there's fewer variables to address. A PSTN/TDM phone call takes a discrete unit of bandwidth per call, either 56K or 64K depending on the underlying transport technology. The only variables are start time and duration. Erlang modeling, based on queue theory, addresses this quite well; it isolates start time by normalizing duration to 3600 seconds/call and provides useful, realistic measurements.
IP networks, though, have difficult to model traffic flows with packets of varying size, varying latency from node to node--and from packet to packet within the flow all transmitted at different start times and with different durations. This is only exacerbated by the variety of applications on the network. Variables are nearly impossible to isolate (practically) and capacity planning is more reliant on utilization trend analysis rather than proactive modeling.
As an example, the network I help operate sells ISP service over DSL lines provided by a local carrier. We have a meager 300 or so customers that have DSL products that range from 384K down to 3M down. Let's normalize all of them to 1M to make the math straightforward. We pay our upstream providers about $30/M each month for connectivity. So you would have me pay ($30/M x 300M)==$9000 per month to support those customers. That's more than I currently charge agreggated across the whole group for DSL service.
Now, in reality, what is the actual average utilization for those 300 customers? Three megabits per second on average for the whole group. And that's just the amount on the direct circuit from the local carrier--not the amount from those customers that use my upstreams. Around 14% of their traffic goes to other customers on my network, so only 2.6M or so actually goes upstream. That's around $75/M monthly on average for upstream. Now I can afford to charge what I do, and still provide email, personal webpages, news, DNS, etc, plus staff 3 tiers of support. BTW, peak utilization for these customers doesn't exceed 5M 99.999% of the time.
Also in reality, I also have thousands of T1 (1.5 Mbps), dozens of DS3 (45 Mbps), six OC3 (155 Mbps) customers and 14 GigE (1000 Mbps) customers. My peak daily upstream utilization is around 800 Mbps for all customers combined. It's never spiked above 924 Mbps, including DoS attacks.
I price and operate my services according to that reality, not magic or fantasy. If you feel that means I lack common sense, then I submit that common sense...isn't.