interconnection costs

Thu Dec 24 04:08:53 UTC 2015

Hi Reza,

There is some terminology and view point confusion... 

First of all, it is Network(s) which connect with Other Networks using AS #, not AS's which connect with each other. (In a particular view (Routing) it could be said that AS's connect with each other).

Second, A connection (interconnection) is just that...as simple as a piece of cable that is connecting two routers together (routers are next to each other)....In a complex example, the two routers can be in different parts of the world.

Today we take Colo Providers for granted, in realty they merely provide " Rent a Data Center" +  "Rent other services on demand".
Originally these folks used to make their money by renting space, power, cooling  etc.. However they have figured out over the years that they can make a lot more money by 'renting' a piece of wire, and lately they trend is to extract as much money they possibly can for renting these wires as they can get.... These are the x-connect charges... 

Other have pointed out that 'peering' comes in many flavors, and it can be argued that IP Transit (aka internet access) is the top of the paid peering interconnection type.

There are a number of 'Eco-systems' in play if you want to look at the whole picture. There are Last Mile network, Middle Mile Networks, Long Haul Networks, Data Center(s), Peering Fabric's etc.

Today connectivity decisions are influenced by Location, Cost of getting to that Location, and Cost of having Presence at that location. Which in turning out to be more and more dictated by Long Haul connectivity providers and mostly by the terms and conditions of the Colo Providers. Older Colo's where lots of major carriers are present, tend to be the most arrogant, expensive and challenging companies to do business with, however their competitors who tend to be far more reasonable and cooperative to work with, are lacking in their list of 'Networks' that one can connect with...thus forcing difficult decisions.

Another interesting thing to note is that Different Markets have Different Colo Folks holding the 'premier' position, and they pretty much all act that way in those markets, the same operator in other markets where they don't have a lead position are much more reasonable to work with, (and there are always some exceptions).

So from a network (ISP) operator's perspective the following appears to be options listed in graduating manner.

        1)  Buy IP Transit from who every is available (Typically a very limited choice if any, and pretty much no negotiating leverage)
        2)  Buy Middle Mile transport to nearest Data Center in Town. (Better options than 1, unless the D.C. is a major one, otherwise still limited options)
        3)  Buy Long Haul to nearest Major Data Center...(Better options then 2, costs get to be interesting).
        4)  Buy Long Haul to nearest Major Data Center and establish a POP at the DC    
        5)  Buy Long Haul to nearest Major Data Center and establish a POP at that DC and extend your connections to other DC for connectivity (because who you want to connect to is not present in the other Data Center).
        6)  Buy Multiple Long Haul to different Major Data Center and establish POP and connectivity etc etc etc.

So after painting this picture... What costs do you want to analyze ? Not everyone is building networks in the same manner..

:)

Faisal Imtiaz
Snappy Internet & Telecom

----- Original Message -----
> From: "Reza Motamedi" <motamedi at cs.uoregon.edu>
> To: "Baldur Norddahl" <baldur.norddahl at gmail.com>
> Cc: "nanog list" <nanog at nanog.org>
> Sent: Wednesday, December 23, 2015 7:39:11 PM
> Subject: Re: interconnection costs

> Aren't availability, guaranteed service and remote hands an incentive to do
> peering inside a third party colocation?
> 
> I see very large numbers for xconnects for instance in Equnix [
> https://blog.equinix.com/2013/08/equinix-cross-connects-hit-110000/] and it
> made me believe buying xconnect is still a normal practice.
> 
> Best Regards
> Reza Motamedi (R.M)
> Graduate Research Fellow
> Oregon Network Research Group
> Computer and Information Science
> University of Oregon
> 
> On Wed, Dec 23, 2015 at 2:12 PM, Baldur Norddahl <baldur.norddahl at gmail.com>
> wrote:
> 
>>
>>
>> On 23 December 2015 at 20:05, Reza Motamedi <motamedi at cs.uoregon.edu>
>> wrote:
>>
>>> In Private peering however the AS pays the colo provider for the xconnect
>>> per ASes that it wants to peer with. The cost of transit would be
>>> additional if the peering is in fact a transit and not settlement free.
>>>
>>
>> You are still assuming there is a colo. But perhaps the most common case
>> is a multihomed company buying transit from two independent service
>> providers. The customer is at his office and the two service providers will
>> have their end somewhere in the city, perhaps even terminating their end of
>> the circuit in a street cabinet. The customer is multihomed and therefore
>> has his own AS. This is a peering situation with three AS numbers that fits
>> your description, it is private peering and there is no xconnect. Instead
>> there is usually a leased line cost, but this cost is often hidden from the
>> customer. Also the ISP might own the line (physical fiber) and the cost is
>> not a simple $/month.
>>
>> But also two ISPs might peer in this way. Residual internet providers have
>> a ton of points of presences, so why choose a place where there is a
>> xconnect fee? We can peer anywhere in the city, including at a random
>> street cabinet. Often the cost of renting a dark fiber somewhere is lower
>> than a xconnect fee (a sign that datacenter owners are too greedy).
>>
>> If one party is a content provider I give you that the peering point is
>> usually at a datacenter somewhere. But still, if the content provider is
>> big enough to run their own datacenter, we are back at the leased line case
>> again. Some content providers, even if small, prefer to just run their own
>> datacenter in the basement of their offices.
>>
>> Regards,
>>
>> Baldur
>>