It is what it is....

Wednesday, January 24, 2007

Peering at the edge and the role of the P2P

Here are some reasons why I am leaning that way:

1. Ubiquity in access and Choice. Everyone, everwhere on the planet and perhaps elsewhere will eventually have the ability to choose who their isp is and how they connect to their isp. ISP is defined as a company who sells transit. How they connect is defined as the last mile and example are hfc(cable), fiber, copper(DSL), wireless(wifi, wimax, satellite, whatever)

2. Networks are only as good as the networks they connect to. The whole point of p2p was to improve performance and save costs, at least that is my impression of how p2p's position themselves commercially. p2p technology works best when the 'sharing' happens locally, as in the same town, region, neighborhood because the data doesn't have to encounter so many hops along its way from one peer to the next.

3. Last mile service providers don't connect with one another. This is stupid on their part! Example: I'm on comcast and my neighbor uses pacbell dsl. I just downloaded the maverics surfing competition video which i got by seeing it in my koniki client. I watch it and move on but it is still on my computer and kontiki is still running in share mode. My next door neighbor fires up his kontiki client and he too sees the link for the surf comp and clicks it. Does my neighbor get that file that is sitting on my machine or does he get it from some other location on the net? That depends on different factors but chances are it isn't coming from my pc. Stupid because this isn't efficient and when services aren't efficient they will be driven to the ground sooner rather than later. It wouldn't be efficient for my neighbor to get to my pc vs some other node on his isp's upstream provider(for example sake pac bell, sbc and att and all separate companies) even though he can see my pc from his home office window because in order to do so it is going to cost pac bell money, it is going to cost comcast money, and it is going to cost the surfin website company money and its going to take a shit load longer and have no QoS. That little request for the video i just watched, should it have to come from my pc to my neighbors, would have the itinerary below:

neighbors pc to pac bell Central office to pac bell regional hub to upstream provider to some public peering switch to Level3(comcast upstream provider) to Comcast regional datacenter to headend to neighborhood node and finally down to my pc. and then head back(next door to me) in some similar fashion....I think there is another hop between the node and headend too so it's even longer has more potential points of failure.

For this reason, it would make sense to move the peering to the edge and the p2p networks are a potentially more efficient way to peer than what is currently done. This could take place in or at the CO level or Headend level by merely interconnecting the two and running bgp or some other protocol that accomplishes the same thing. In this example, it would save comcast money that they pay to L3 for their upstream connectivity, it would save PacBell money that they pay to uunet for upstream, it would save the peering provider money because the router would consume less power and generate less heat and not be taxed by the demand of the reques. ie, it's more efficient. The old school telco boys with their closed networks and fear of new techonologies and competition are slowly but surely retiring and moving on because they've been at it for 30+ years and they have better things to do like play golf and play golf again. Backfilling for them are leaders from the IP generation and their mindset is more progressive, they understand the layer one is always going to be layer one and that it is a commodity. As such, they aren't as concerned with keeping it closed at the local level as their predecessors once were. They see the value and understand the threat of not opening it up. Its better to get something than go out of business right? They're keenly aware that their incumbant status could be a liability on several fronts including net neutrality supporters stirring up the pot and filing lawsuits, new technologies emerging which are more efficient and have low barriers to market entry and they still owe a shit load of money to pay back the banks for paying for all that fiber so they wise up and collectively develop a plan to have some arbitrator transfer packets amongst themselves. The arbitrator has to be a neutral third party to avoid any conflicts of interest between the networks and has to have deep reach into the networks users hardware and obviously know what they are doing from a technical perspective.

A couple p2p providers, specifically kontiki, are in a great position to jump on this. It's no wonder they were purchased by Verisign!

Labels: , , , , , , , , , , , , ,

3 Comments:

Anonymous Anonymous said...

pushing peering to the edge? hmm... have you thought about the impact on traffic flow over the internet? Not to mention the political obstacles you need to overcome within each carrier to do this.

8:29 PM

 
Blogger tomo said...

anonymous,

thanks for the note. yes, my point was that for most networks, their routing rules are to dump the traffic as soon as possible. in the past this has been what happened at the tier one backbones because the LECs didn't have backbones, they were at the transport layer. that has changed and what we used to rely upon the LECs to provide can now be provide by the LEC, CLEC, Cable Co, Satellite Co, Wireless Co, etc. this presents the consumer with choice of what last mile medium they use and what provider they get that from. what is different now is that all of these last mile providers have national backbones with adequate capacity and all the control they want, as long as that packet stays on their network. the issue is that exchanging that traffic with a peer(ie, you have an open peering relationship at an exchange point like eqix, paix, mae', etc) likely adds latency, is not the most efficient in terms of cost(isp's pay for capacity too, both at the peering points and to their upstream provider or wavelength/fiber provider and not having to grow that capacity means not having to spend money to upgrade it. local peering, imho, is a savior to the LEC market as we know it today because if peering is taking place on it there is a big value and the propensity to move away from something that works and saves money is low when compared to the propensity of innovation as it relates to breeding disruptive technologies that, at the end of the day, are more efficient than what they replace. wrt to the political obstacles, this is kinda happening now on the content side with companies like yahoo, google, youtube, ebay and others who peer on local exchanges like the ones equinix runs in some of their sites. those aforementioned content companies, for lack of a better classification, peer with each other and with many many tier two and tier three isps. that was equinix' whole sales pitch with their gig-exchange service: Equinix Sales Guy;

"Mr Yahoo you know you are paying transit costs to connect to Mr Hurricane Electric in the cage next door. Mr HE, you too are paying transit costs for your traffic to get to yahoo in the cage next to you. Why don't you too just run a fiber between each other and peer. That way you bypass all the transit costs and have greater control and scale of your interconnectivity. Wait, Mr HE, did you just say you would love to be able to do the same thing with google, ebay, youtube, EA, etc. etc.? Perhaps you may be better off connecting into equinix' exchange switching fabric via a gig port that can be subnetted and provide you with a single connection that will aggregate all of your peering traffic to a single port run over a single cross connection from your router to equinix' switch. Cost? Yes, there is a cost for the port on the equinix switch but the point is that you can't afford not to do it, it's is single digit equivalent in $ per meg and give you the control and scale that large networks enjoy."

11:12 PM

 
Blogger Ant Bryan said...

you might want to look at metalink ( http://www.metalinker.org/ ). it's working on some of these things.

7:30 AM

 

Post a Comment

Subscribe to Post Comments [Atom]

<< Home