Slashdot Mirror
Bandwidth as Commodity
TwoSticks writes "This NY Times article (CT:required
annoying but free registration) and one at Yahoo talk about the Enron corporation working to set up a market for bandwidth, similar to the existing markets for buying and selling gas and electricity, but with transaction times in seconds. They claim this is essential for next generation network management, and I'm inclined to agree. "
I think you and i agree about the lack of similarity between a natural gas pipeline and Internet bandwidth. My point in making the comparison was to demosntrate that the two are not the same. ENRON is a company whose primary business is moving natural gas in pipelines. Obviously, if they are proposing this "network bandwidth as a commodity" model, it's because *they* think they can "leverage their core business," and apply similar principles to Internet data. (I don't agree)
For technical reasons, the top-tier ISPs have long been reducing the number of other ISPs with whom they "peer" or exchange data. The complexity of the routing tables were becoming the biggest impediment to timely transmission of packets. At the backbone level, the top-tier ISPs want (and to a large extent have) routing tables that are simple, and cover very large ranges of IP address.
For example, they want to route all of 204.0.0.0 to one particular destination. What they hate is "top-level" routing of small ranges (like a Class C: 204.2.12.0) to a particular destination. This consolidation has very clearly pushed almost all but the biggest ISPs out of the top-tier.
Let's look at this from a small regional ISP's point of view: "Joe's ISP." Joe doesn't have the heft to make a peering agreement with each of the top-tier providers, so instead, he signs on with one, like UUNET. UUNET sells Joe some class C or B netblocks for Joe to use or allocate. These are specific IP addresses that are routed on the 'Net at large because of peering agreements that UUNET has established with the other top-tier providers.
As far as all other top-tier providers care, traffic to Joe is really traffic to UUNET.
Joe cannot sudddenly decide to buy 1 GB of bandwidth from someone else, because he would then need new IP addresses for all his machines, new Internet DNS entries, a new T-1 connection (at least a month from the local Bell) to another top-tier ISP, etc., to make it all work. Ask any (TCP/IP) network admin what it thinks about changing every IP address in their domain once a quarter and you'll get a very clear answer. (But i'd make sure they weren't holding anything that's going to hurt when they throw it at you.) Changing IP addresses is something that any network admin with a brain in its head *never wants to do.*
So if Joe is effectively eliminated as a purchaser of this network-bandwidth commodity, the only people who would be in the market for it are the companies that already control the market. WorldComm-UUNET, Sprint, AT&T, C&W and the other top-tier ISPs don't need ENRON to step in and start brokering the data between them. When they get the technical means to charge by the packet, or dynamically route data by the cheapest ($) path, they'll do it *by themselves*.
For a long time, I thought that what the web really needed was a Differential-HTTP. This makes particular sense with a site like slashdot. I punch in every few hours, but let's face it: 90% of the page is the same. At most, there's a new blurb or two at the top of the page.
/. is cool enough to generate dynaic pages, but think of the poor IIS admins :)
The existing standard is horribly inefficient from a bandwidth perspective. It makes a lot of sense, though, from a server load perspective. It's usually easier on a machine to serve the same page 10,000 times than it is to serve 10,000 different updates. (Okay, so
Right now, the web is designed with the assumption that an extra megabyte is free, but an extra CPU cycle isn't, at least, not at the big sites. If you read the HTTP specs, it seems that most of the recent work is designed to shift CPU load from the server to the client.
Even though end users probably won't see a difference (Let's face it, tracking usage on a modem, or even a 10Mbps DSL line isn't worth the paper and trouble -- this is strictly a game for the serious bandwidth consumers.), we're going to see a sudden shift toward minimizing bandwidth from the standards organizations. This might have a negative impact on the hardware budgets of the larger web sites.
Forward, retransmit, or republish anything I say here. Just don't misquote me.
Data Communications ran a really good, comprehensive article on this in this month's issue. It talks about the competitors in this market, strengths and weaknesses, and issues like fraud. I think Enron's PR department is just doing a better job than those of folks like arbinet, which is why they get credit for 'proposing' an idea that's already been implemented by at leat seven companies.
FYI, the article at Data.com says you can get a T1 from NY to LA for between $3900 and $4800 through the brokers, compared to three times that from major carriers.
Data Comm...gude sctuff.
1. Fixed Costs (local loops)
2. Variable Costs (switch capacity, trunks)
3. Billing Costs (counting packets/calls)
The problem with metered service is that the price for call minutes or packets is usually far in excess of the true variable costs. Phone companies want to price products on "value", not cost plus reasonable rate of return.
Another problem is the high cost of the infrastructure needed to collect usage information and generate bills. This is a big number for circuit switched voice, I suspect it would be worse for packetized data. It can be cheaper to provide a flat rate pipe than add all the cruft required for metered service.
The last time I saw some numbers, about 70% of the cost in providing local phone service was in non-usage sensitive things like providing and maintaining the subscriber's local loop. The prices charged by phone companies and PTTs have very little to do with the costs of providing service.
Brokering bandwidth isn't something that small shops and users will have to deal with. The power companies deal bandwidth at the lowest level. I am sure that only NAP's and other large AS's will find this useful.
Sure, a network of caching web proxy servers sounds OK. But there are issues...
would SlashDot be served well for users on @Home who sucked up the recent "reconfigure your browser" e-mail (i.e., to configure it to run againt @Home's proxy servers)? Sure, they seem to get things downloaded faster, but what about real content?
What about more nefarious things like "intelligent, filtering" proxy servers that sort of lose "bad", "inappropriate", etc. data?
Maybe if HTTP could be split, or maybe when things like CSS, etc., could allow two streams of download: one stream for graphics, the other for the text. But none of the Ad-driven sites will go for that because it'll mean it's too easy then to just ignore the graphics pipe...
Then there's service level agreements, what happens during a fiber cut... it all sounds like a nightmare. I think it is a neat idea, but the devil is in the details on this one.
But this may not be so good for the everday users. We're back on the cost-per-connection model that everyone loves to hate.
First, we had phone companies that controlled everything, then they were broken up into regional ones. Then we had deregulation and competition for local and long distance carriers. Now we have a whole mess with xDSL where you can sometimes end up making separate contracts with your phone company and your ISP. The xDSL service providers can shop around for the cheapest use of the network, and customers can shop around for the cheapest service provider and the cheapest local phone carrier.
The ultimate goal seems to be more competition, more customer choice, and cheaper service. I see this "bandwidth commodity" idea as simply continuing the trend to it's ultimate end. We (the end users) get more options and cheaper rates, with admittedly, more confusion. But in the end, we come out ahead. I say go for it, take deregulation to it's logical end; that's what this seems to be.
Every bandwidth consumer must be connected to a "pooling point" at both ends. A pooling point is a router or group of routers, and all bandwidth sold in the commodity market is between pooling points.
When the consumer reaches an agreement with a provider, both parties contact the pooling point operator and receive a /30 range of IP addresses. (That is, 4 of them.) If the bandwidth changes hands, the IP addresses are assigned to the party that receives it.
Then, when it's time to actually deliver the bandwidth, the routers must be configured to send traffic like this: (The last two bits are not actually specified in the document, just here as an example)
Consumer (source)
|
| local, "fixed" network (leased line?)
V
IP xxxxxxx00
San Jose pool
IP xxxxxxx01
|
| provider network
V
IP xxxxxxx10
Wash, DC. pool
IP xxxxxxx11
|
| local, "fixed" network
V
Consumer (destination)
So, what the bandwidth provider is selling is a promise to route traffic coming out of IP xxxx01 through IP xxxx10 (and the other way around, presumably), and the consumer must direct all traffic it wants to go on that link through the corresponding IP addresses. In other words, the consumer treats the "middle section" like a router with connections to both of his sites.
The problem I see with this is it makes routing tables huge. Every connection you sell needs a separate routing entry, since they're unlikely to be aggregateable. Plus, you'd have to either have a dedicated line between the two ends you provided, or do tunnelling, to guarantee that you treated the packets as the contract specified. (Otherwise, how can you distinguish between packets on your network from one contract and another? Remember, the original contractee can resell the bandwidth to anybody, so you don't know what the actual source and destination addresses are until the packets actually start flowing.) The bandwidth provider has to do routing based on the gateway address, not the source and destination.
Yes, but the difference is that your upstream provider is always the same one. The point of bandwidth brokerage is that you have several upstream providers, each bidding to get your traffic, and you choose where to send each packet based on cost, but also quality of service. You can also offer transit capacity to two ISPs which are connected to you but not to each other. This is of course already done, but usually by way of writing new peering agreements, signing them and returning them by fax, i.e. slow. A bandwidth brokerage would speed this process up and also offer a measure of automatability (no humans involved). And then of course, you can do all the things you can do in real commodity markets: futures, options and all the other derivative thingies that I don't understand a thing about. :)
Anyway, that's my take on bandwidth brokerage. The article might of course be talking about something completely different.
Any technology which is distinguishable from magic is not sufficiently advanced.
Bandwidth supply is increasing at a rather brisk rate. Unlike gas, and most other commodities, you can't just store it up and put it on the market at a later date. I think their drawing of analogies are kind of suspicious.
I don't think bandwidth really qualifies as a commodity. This whole idea is pretty scary.
brian
I think you're thinking in the wrong direction. Comparing the internet with gas lines is just a bad comparison, because, indeed as you say, on one side, a cubic foot of gas has an intrinsic value, while a kilobyte of packets doesn't. But on the other side, bandwidth can be shared between dozens of consumers and providers. Try sending one specific cubic foot of gas from one specific point to another specific point through that big net of pipes... :-)
Comparing gas pipelines and the internet is like comparing melons and oranges. They're kind of similar, but the similarity is only superficial.
The internet can much better be compared with mail and transport. A pound of mail has no intrinsic value either, nor does a cubic foot of mail package, just like a kilobyte of packets. One cubic foot of mail package can travel seemingly independently from one consumer to another through the mail system, just like a kilobyte of packets can do that through the internet.
How does the mail work? You pay for upstream mail only, and many companies, as a service, give you postage paid enveloppes for your return mail. The internet could work the same way, and that's what I think the general geste here is.
But then still there's the bit about the bandwidth. Basically to you nothing would change. You'd just pop your packet into the mailbox, and your ISP would empty that mailbox. But then your ISP has the choice to send it on with UPS, with TNT, with American Express, and so on, and your ISP would choose the best option regarding speed and cost. You, the consumer, wouldn't notice very much of a difference, except maybe a cost decrease and/or speed increase because your ISP can easily choose what's best.
)O(
the Gods have a sense of humor,
Never underestimate the power of stupidity
To err is human, to moo bovine
First of all, I don't believe the infrastructure to buy and sell bandwidth is in place yet, at least for arbitrarily located machines. This is being worked on, for sure, but in general case I don't think you cannot buy or sell anything but what is essentially a leased line.
You're thinking of the n^2 possible connections between the sites on the network, and if you did need to connect them all, then yes, it wouldn't be feasible to trade bandwidth like a commodity. Fortunately, this is not the case.
Recall that the basic network topography is more or less hierarchical. (If you imagine it as a balanced binary tree, then the number of hops to connect two arbitrary machines is O(log n), where n is the number of nodes in the network, and there's a 50% chance that you will pass through the top level on any given connection.) So you can start with bandwidth trading on the routes between the big backbone sites and still reap much of the huge benefits of commodification -- after all, almost all traffic passes through the backbone eventually. (For those who've forgotten their econ: the benefit of an exchange is that it becomes much easier to match up buyers and sellers. Think how much aggravation and time would be saved if, say, houses could be bought and sold in minutes rather than weeks.)
And as the market matures, bandwidth trading will naturally spread to lower and lower levels of the network tree. I think it's unlikely that you or I will be trading bandwidth any time soon, though, any more than you or I invest in oil futures before heading to the gas station, but odds are that very soon our ISPs will.
Besides, there is a book (IIRC called "Virtual City", but I am not sure) that very well describes a similar system where you buy remote processing power in micro-chunks on the as-needed basis. The book is recommended, by the way, it explores the consequences of being able to transfer human consciousness into a piece of software.
You're thinking of Greg Egan's Permutation City, which is lots of fun if you're into AI, madness and Platonism. The ending is seriously weak, though.
Bandwidth bartering is a little silly today, given the high cost and low speed of residential local loops. But imagine how things will change if local loop / MAN connectivity (Metropolitan Area Network) starts resembling a LAN more than a WAN (fast, ubiquitous and cheap).
First off, it's likely that your average person will have a much higher local connectivity speed than they could expect to use for free over the WAN. This is not the case today- the Internet WAN is SO much faster than most people's MAN connection (their analog POTS modem) that there's a reasonable expectation that you can get a MAN speed connection between any two points on the Internet WAN without paying extra. This will not be the case as the multimegabit MAN becomes a reality. I can't expect to get a 2Mb/s connection from my home in Houston to my Dad in Chicago for "free".
However, both producers and consumers of content and network-based applications will want to be able to take advantage of that new MAN speed, and are likely to be willing to pay for it. This creates a market for guaranteed quality bandwidth from point to point. In the consumer market, this need is likely to be on-demand. If I decide I want to "rent" a streamed episode of the Simpsons from the Fox website, I'll be happy to pay a few extra cents for a guaranteed quality connection for the time I need it to watch the show. This won't negate the need an unguaranteed Internet like we have today, but will create the financial motivation to introduce guaranteed QOS either through separate parallel networks or some sort of traffic prioritization. I believe Enron is building a parallel network.
This dramatically changes the economics of the WAN bandwidth. Right now, both information distributors and consumers share the cost of WAN connectivity by paying to connect to an "upstream" Internet provider. The producer pays it directly, and the consumer pays it as part of their ISP bill. However, in a world where high quality bandwidth is requested and paid for on demand, this cost burden must shift to the producer entirely since it's the producer who is providing the on-demand service and will be able to pass the costs of that on-demand bandwidth to the end user directly (or subsidize it with advertising).
As the guaranteed QOS WAN bandwidth companies begin to compete for the business of the producers, it becomes critical that they be able to promise nearly universal reach; to be able to connect to as many consumers as possible. So they will be motivated to connect to as many of the local loop providers (local ISPs) as possible. This won't happen if the local loop provider has to pay for the WAN connections, so companies like Enron will offer connectivity to ISPs for free or even pay them for the right to reach their subscribers.
So in the new world, it works like this. A content or application distributor will connect to some sort of MAE-like bandwidth trading facility and buy WAN connectivity from the various WAN vendors located there. At first, the bandwidth units will be traded in large chunks, but it will become more and more granular as routing protocols evolve and dynamic financial-cost routing and accounting becomes more practical. The content producer will either charge the customer for the content and delivery bandwidth, or will pay for the services via advertising. The consumer's ISP will be paid by these WAN bandwidth companies for the privilege of reaching the consumer, who may end up paying nothing for their MAN connection (similar to the TV broadcast model).
Enron is betting the farm that this will happen. They're investing big bucks in building this huge fiber network in anticipation of providing this service, and they are already hooking up to local ISPs to make it possible.
At least, this is how I read it. What do y'all think?
IMHO, network admins are a completely different bunch of people than stock brokers. Who will do the dealing? Will network admins have to learn to phone with 3 people at once? Or will stock brokers learn about BGP4? Only time will tell...
Any technology which is distinguishable from magic is not sufficiently advanced.
A lot of people have commented that this doesn't make much sense, for various reasons. Essentially the biggest argument has been that it's not technically possible to do this for arbitrarily located systems.
However, in most cases you buy a circut from a telco (e.g. USWest) which is then connected to an ISP. You can buy the circut from the telco without setting it up to connect to anything and it would be useless. In this way bandwidth can be seperated from where it travels much as electricity is not the wire it travels across.
Many ISPs maintain much bigger circuts with telco's than what they are actually using in bandwidth at any given time. And many maintain lines to multiple backbone providers (e.g. UUNet, Sprint, etc...).
As such bandwidth could be shifted from one backbone provider or additional service added as desired. Obviously the technology isn't ideal because it can't be done automatically based off a commodity market that knows how much delivering a given packet will cost in monetary amounts. But it can be done on a much more crude hand basis.
This isn't going to be useful for smaller companies or even individuals. But that's not the point. Just as smaller companies and individuals don't go to Enron to purchase their electricity for their houses, nor would they purchase bandwidth.
It's still early on in this idea and it's still developing but the idea does make sense and it is possible.
In the UK the academic network charges the universities for transatlantic bandwidth.
http://bill.ja.net/common/faq.html
However use of the provided caches is not charged in this way. Net result was a drastic increase of cache usage by universities. Far more efficient usage of bandwidth, people paying for their share but still a long way off being a commodity to be exchanged yet.
This is one of my areas of economics . . .
.. .
There is fundamental differences between this and other commodities markets, even the newer ones such as electrical distribution. Electricty or gas, once you get it there, is the same regardless of where it came from. Bandwidth, on the other hand, is the "getting there" portion of this. Electricity *can* be moved from Boston to San Francisco, and so can gas, by bumping the gas along the way (though it's not the same in & out).
But bandwidth is always betwen A&B, and the only way to replace this is A->C->g->B. If you don't have ends at A & B, you just can't do it.
Some type of commoditization is certainly possible, but I've alwasy assumed it to be through a semi-intelligent part of the packet that chooses a path as it goes based on the cost& lag of each path
The chronological problem isn't to bad; it's been solved for electricity--you buy howevermany gigawatts for 2 pm on thursday, just as bandwidth would be. But there's those two loose ends of the path that need to be dealt with for bandwidth .
I may be missing something, but it is my impression that we are talking here about what's called "reserved" bandwidth. For example, let's say you want to videoconference between New York, London and Hong Kong over IP networks. If the route of your packets takes them through a heavily-used segment (think Starr report downloading), your videoconference is hosed. The idea is that you would be able to reserve bandwidth, which will be guaranteed to you, as in "you are guaranteed to have sustained 1Mb/sec speed between machines foo.com and bar.org from 15:00 to 16:00 on May 24, 1999". AFAIK this is currently technically infeasible for randomly located machines, but is heavily worked on.
Kaa
Kaa
Kaa's Law: In any sufficiently large group of people most are idiots.
First of all, I don't believe the infrastructure to buy and sell bandwidth is in place yet, at least for arbitrarily located machines. This is being worked on, for sure, but in general case I don't think you cannot buy or sell anything but what is essentially a leased line.
Second, for this to work well, we need some kind of micropayments structure in place, plus reasonably intelligent software agents that would be able to go out onto the net and buy bandwidth for us when we need it. I don't see this happening in the near future. In five years we'll see.
Besides, there is a book (IIRC called "Virtual City", but I am not sure) that very well describes a similar system where you buy remote processing power in micro-chunks on the as-needed basis. The book is recommended, by the way, it explores the consequences of being able to transfer human consciousness into a piece of software.
Kaa
Kaa
Kaa's Law: In any sufficiently large group of people most are idiots.
Bandwidth as a tradable commodity has got to be
one of the most evil "Internet" money-grabs i've seen come down the pike since 1993. Think about it: when you add in a middle-man who make his living by trading what Party A creates and Party B consumes the price goes UP, not down. Look at any manufacturer-distributor-retailer model.
The next problem with this is that Fortune 1000 corps. are not going to accept a "Sorry, we couldn't buy (or sell you) any more bandwidth today" explanation from their ISP or net-broker. So you can bet corporations will be buying their own big chunks from their ISPs. Which means that only the little guys are going to feel the price difference here.
The third problem with this is that network packets are not natural gas. Natural gas is purchased by corporations like ENRON at THE SOURCE, and then transported to a refinery. This model works because a cubic foot of gas (or gallon of crude) has an intrinsic value. Your data is usually worthless, except to you and the person you want to read it.
This is a really good idea to support if you want to be paying 5 times what you are now to some talentless fool in Chicago or New York who's raking it in today because some other idiot put a back hoe through the piece of fiber your packets usually blink across. And if you want inferior service.
By making bandwidth a commodity that can be traded and bartered, we can start using it efficiently.
Currently, most (not all) web surfing could be adequately serviced by an extensive network of caching servers. Currently, most users get content directly from a site's main servers, which most likely is not the most efficient means of distributing data.
Also, compression techniques will improve as vendors attempt to be more miserly about the bandwidth they use - most web content could be compressed for transmission much more than it is now.
Compression and caching systems are but two technologies that would improve dramatically if we had to treat bandwidth as a commodity.
Look, there is no reason why most of us should pay anyone for bandwidth. That's just a myth perpetuated by telecoms companies.
:v)
All we need to do is turn the world into one big, anarchistic, wireless WAN and run it by the people, for the people.
See http://www.indranet.co.nz for an 11Mbit system currently trialing in New Zealand and Freenet on http://www.dcs.ed.ac.uk/~iic/4yp/
The sooner people get off their arses and organise this, the sooner we'll be free of the complexities of ISPs, large IP bills, government censorship and dependence on telcos. All it needs is for someone to make an affordable comms link that will work across a road, and to make the design open source. Let's face it; that shouldn't cost more than a modem does now.
Vik
PS If you have such a device LET ME KNOW!
Here's a link to it on amazon.
Very interesting ideas about all sorts of things: auction markets in processor time, continually improving AI spam and spam-filter contests, and some of the philosophical and technological consequences of transfering consciousness into software.
Not to mention a good story.
The enron.net web site has a white paper, too, but you need to register to download it.