I was wondering the importance of content to IP providers. Is it feasible to
go after a lot of hosting companies and such as a business model and greatly
skew your traffic ratios to hopefully reach a critical mass. I would think
at some point you would have so much content that people would start to come
to you for peering or to purchase access to get to that content which would
cause a reduction in overall transit costs, but what would that critical
mass be and how valid is that thought?
Unbalanced traffic can cause difficulties with peering. The eyeball
heavy networks will tend to peer with you but a long list of large
(route table) players will not.
Be careful with this approach to gaining peering.......While you may gain
some, you will probably end up paying more in monthly transit fees than its
worth. Speaking from experience here.......Having worked for a national
player that took this approach (prior to my involvement with the company, so
I did what I could with what I was given, thanks to an acquisition), I don't
care how much content you have, if you can't prove that you can meet the
remaining requirements that peers set forth (ratios, network size, # of
pinball machines per node), you aren't getting the peering. Most of the
"large" networks don't care about your content. They know if you fail to
provide quality routes to your customers, your customers will eventually
give up (I don't care if you only charge them $10/mb) and become their
customers (or customers of someone whom they already have a significant
peering relationship with). Ideally, if you wanted to obtain quality
peering, you would focus on gaining both types of traffic and sustaining a
"reasonable balance" of push:pull. If not........I hope you have a lot of
cash in the bank to continue paying VERY high transit fees to accommodate
the volume of content that you bring on (Hopefully you charge your customers
more for transit than the rate you pay your upstream and you'll be OK ;).
If you fail to do so, you should see the implosion coming a mile away. All
of this being said, don't build your business model around gaining content
to gain peering, b/c these days, there is no such thing as critical mass.
Just my $0.02
Chicken and the egg.
About 1997 - early 1998 time frame, there was talk of some big content players getting together to do something similar. I remember it was some large financial institutes, Disney, etc. There were 5 serious content players that could not get peering and were being forced into a corner to pay what they considered extreme pricing for transit. They were all in Pacbell territory and the talk was coming together to force peering with some of the larger backbones or having them lose access to the content.
It was a good plan except you have to be willing to following through with the painful poker game. There could be an outage of about a week before a certain backbone might cave in. Still the plan would see to have a good chance of succeeding as long as you went 1 backbone at a time. That way you do not create an organized opposition. Most downstream users would complain loudly if they could not reach their brokerage accounts, or yahoo, or perhaps MSN etc etc.
With many very large content players building out new backbones now, we may see some of that. More likely it will be pricing pressure on transit, not transit-free peering.
Scott makes some good points. (How is that shinny corvette?)
Shane as far as the "thought" that backbones will "pay" to get to your content. It's just not going to happen. If the content were that important they might go directly to your customers and offer them a wonderful deal to buy a link from them. Or they might "give" them peering. Or with much of the key content now residing of servers like akamai, how much could u really make per meg since akamai resides servers across different backbones. It's an interesting poker play, a long shot. Still if you got yahoo and MSN to go along with it.... then again you still have to argue the upside to working through a 3rd party.
The reason it's not going to happen is that in today's economy it's more important for a content provider to have their particular content delivered to the end user than the end user wants the particular content. Most end users have choice, they can get the same or similar content from many different sources. Example: they can get news from CNN, on MSNBC, or the Washington Times, or the NY Times, or Reuters, etc. Most content providers need every eyeball they can get, they can't afford to choose who can or can't see their content. Thus, if there's a temporary or persistent connection problem between the two parties (content provider, end user's eyeballs), in the aggregate it ends up hurting the content provider more because in most cases the end user will usually just go find the same or similar content elsewhere. CNN learned this lesson on 9/11. When end users were unable to reach CNN, they went to other sites. This hurt CNN a lot more than it hurt the end users.
If the problem is with an "important" connection, say between a user and their online brokerage account, they will either change IPSs or change brokerages. Which is easier to change? For most users, it's easier to change online brokerages.
That's why Akamai has a business model, content providers pay extra to ensure that their content can get to the end user, and get there fast, so that the user doesn't turn away and go to another site. That's also why "user pays" websites haven't been very successful, except for niche markets (where the content is specific to the end user's needs, and not widely available elsewhere online).
It is my opinion that eventually the Internet will be mostly funded by those who send packets, and will be mostly free for those receiving said packets, much in the way that 800 numbers are funded in the telephone system. In order for that to work, we will need a settlement system. I predict that something like this will start happening before 12/2005. Certain services that are highly desired and high bandwidth (streaming radio comes to mind) will be funded with a subscription model, so that the end user continues to get the content without paying extra "delivery fees" to the ISP, but with payment to the originating site, and then settlements to the systems that carry packets. Ultimately, it will be free to get packets, and expensive to send them. End users will continue to pay a modest fee to get connected, but will have high speed connections (where feasible) and a modest outbound packet allotment. If they want to send more packets than a typical end user (sending lots of email, or hosting a web server at home), they will also be a "content provider" and have to pay more for the packets they send.
Date: Thu, 11 Jul 2002 08:37:01 -0700
From: JC Dill
It is my opinion that eventually the Internet will be mostly
funded by those who send packets, and will be mostly free for
those receiving said packets, much in the way that 800
numbers are funded in the telephone system. In order for
that to work, we will need a settlement system. I predict
that something like this will start happening before
Agreed, except is any fancy settlement really needed? I predict
Consider peering ratios. I know of at least one up-and-coming
provider (currently in SJC, IAD, and NYC as I recall) that gives
free core-to-edge bandwidth in a effort to meet ingress:egress
quotas. Some established providers will entertain asymmetric
Content hosting can be moved to large cities. Easy. Saves
money. Many people do it. Viewers cannot be herded into the big
cities. It seems that eyeballs are more distributed, and content
is more localized to the large MSAs.
We therefore have asymmetric traffic flows, with more leaving
large metro areas. Surplus edge-to-core bandwidth means cheaper
edge-to-core bandwidth, unless fiber/wavelengths get provisioned
asymmetrically to model the traffic flows.
From a political standpoint, it's easier to get a business to pay
an extra <x> per month to deliver their content than it is to get
Joe Public to pay an extra $10/mo for premium access.
I think 12/2005 is conservative. Industry fallout begins
settling, and we see interesting tactics from "new breed"
upstarts run by laid-off engineers with a bit of money and a
bunch of clue. Beginning of 2005, significant presence by the
end of the year.
It is my opinion that eventually the Internet will be mostly funded by
those who send packets, and will be mostly free for those receiving said
packets, much in the way that 800 numbers are funded in the telephone
Whether or not this happens by economic darwinism through content networks
building out or paying for infrastructure is completely separate from your
next point, which has a major problem.
In order for that to work, we will need a settlement system. I
predict that something like this will start happening before
12/2005. Certain services that are highly desired and high bandwidth
(streaming radio comes to mind) will be funded with a subscription model,
so that the end user continues to get the content without paying extra
"delivery fees" to the ISP, but with payment to the originating site, and
then settlements to the systems that carry packets. Ultimately, it will be
free to get packets, and expensive to send them.
This incentivizes usage and creates an artificial revenue stream.
Simply put, any settlement system in which either party can be paid, will
result in both of the parties causing the type of traffic that results in
them getting money from the network so foolish as to agree to such terms.
If you get paid to suck data then you could pay google to host on your
network and laugh all the way to the bank, you could even pay them extra
if they index the entire Internet every 12 hours, that way the search
database would be real fresh of course, nudge, nudge, wink, wink.
If you get paid to send data, well... that is easy.
See the nanog archives for more examples of why settlement based peering
doesn't work. There was long thread about this two years or so ago.
+------------------- H U R R I C A N E - E L E C T R I C -------------------+
LOL, the subject was just for amusement since I know I'll get alot of
Of course paying for delivery of packets works... settlement based peering
is what doesn't work.
ps. sorry, long week and a too private sense of humor.
+------------------- H U R R I C A N E - E L E C T R I C -------------------+