Pondering many things. Namely, the evolution of the Internet. Without respect for the many NAPs popping up (which is good), without respect for the lack of available bandwidth (which is bad), I am concerned about the evolution of our business. I'll state the obvious to establish a direction. In the beginning there were researchers, and they were good. They built a network, and it was extended to include core groups, and they also found it to be terribly useful. This was extended to Beta in the form of the NSFNet, managed by our friends at ANS and Merit, which gave rise to my neck of the woods, MIDnet, which provided me a passion and livelihood. Off goes the funding, off goes the backbone topology. Enter NAPs, and peering problems. Version 1.0. Oh joy. In the past, we were (excuse the metcalfesque) centrally managed, and issues of peering were resolved by one body. As we examine peering, we happen upon an inquisition into desire, and motivation. NSPs [Network Service Providers] (defined as anyone who is paid to connect some entity to another entity) found that they could go to a NAP or Meet Point, and gain access to everyone else. I can think of smallish turf battles, the CIX silliness, and the quarrels over what it really meant to be "on the Internet". At one point, it meant just to BE at the NAPs, as being at the NAPs implied that you could talk to everyone there. And then, just like Babylon, people stopped talking to each other. And so I view the future, unless things change. That's not to say I'm armageddonish, or even diluvian, just observant of the significant dynamic changes this is currently wreaking. So, WHY would an NSP enter into a peering agreement with another person? Why, to profit from the one side of the connection, to enable an entity [labeled A] to talk with some other entity [labeled B]. In most cases, NSP1 had as customer A, and NSP2 had as customer B, and obv. it was in their best interest to meet somewhere to talk to each other. NSP1 added value to A, by providing a path to B. NSP2 added value to customer B by providing a path to A. So, comes my curiosity, and my puzzling thoughts about the current state of the net. Why is it not in my best interest to talk to NSPX at a meet point? Why, when it is in MY customer's best interest to talk to EVERYONE, would I not converse, and share knowledge and invitations about my customer base? My thought, conspiratorally, is that larger folks (NSP4) could care less about talking to NSP3's p% of the net, when there is a lower cost involved in talking to NSP4's q% of the net, assuming q >>> p. And yet, I fail to grasp why it is not in their best interest to still include that group located in p%, NSP3's customers. Perhaps because they'd rather have the customers? I'd appreciate hearing the rounds of explanation why larger NSPs don't want to talk w/ smaller NSPs. Just because someone has 30% of the internet, they still have an interest in connecting their 30% of the net to .1% of the net, no? Perhaps the geographic cost investment in transit to far-reaching customers is sufficient. Somehow that doesn't answer the question for me. I'm not talking about transit, I don't think it's necessarily in NSP3's interest to carry NSP4's traffic to NSP1. But NSP3-NSP4 I can see as beneficial, w/ no dalliance. -alan "baring my heart for the wrath of all"
Allan laments, summarizes and asks, perhaps rhetorically: { a good summary condensed }
So, comes my curiosity, and my puzzling thoughts about the current state of the net. Why is it not in my best interest to talk to NSPX at a meet point? Why, when it is in MY customer's best interest to talk to EVERYONE, would I not converse, and share knowledge and invitations about my customer base?
{ more good thoughts condensed } Many can answer better than I, and please do not let my world-view restrain other good responses to this interesting human question..... Here is my 'weak attempt' at the human and social side of the issue: If we examine the Kleinrock model for 'large Internets', most conclude that hierarchical routing, broadly defined, is the 'mathematically correct', large internet, paradigm to emulate or to 'grow towards'. Hierarchical routing, unfortunately, translates (in business) to 'big fish at the top' of the hierarchy and 'little fish' at the bottom in 'Global Routing Darwinism' as the net evolves. How does a flat model where 'everybody is free to peer with everyone' move to a ordered, structured 'hierarchy'? My guess ... by attempting to set up NAPS with 'rules that tend to create a routing hierarchy'. These 'rules', written or unwritten, exist; and those in the provider business, by the nature of business, struggle to 'survive' in the 'routing hierarchy'; many wish to 'move up', others struggle to exist, some are 'displaced lower' in the hierarchy by those 'moving up'. Quintessential to the dynamics of business and business relationships is (1) money and (2) power. Seductive and self destructive, yes; but realistically, name a 'technology for the masses' that has changed the dynamics of human behavior for 'the better'. Nuclear power? Broadcast television? Radio? The Press? ..... ?! The Internet? Yes, we communicate faster and faster, but do we communicate better? Does this 'new form of communication' better the economic conditions of the poor or the social condition of the oppressed? I share in the posters lamentation that the 'Commercialization of the Internet' is 'not pretty nor poetic and esoteric'. It has become, obviously, "Commercial" and "Worldly". To end, please allow me to quote from Umberto Eco, "The Name of the Rose", " It is only petty men who seem normal." - Eco My apologies to the erudite for the pedantic, non-technical lament. The technical answer is hidden in the allocortex and other lower brain functions below the neo-cortex in all humans. Greed, desire, fear, self-preservation and superstition to name a few 'core engines' of the lower brain. All the 'mathematics and code-cutting in the human reticulum' are mastered by the persuasion and influence of these little organs just above the reticular formation in the upper brain stem. Life in the NAPS is certainly no exception :-) Best Regards to All, Tim +--------------------------------------------------------------------------+ | Tim Bass | | | Network Systems Engineer | "Never underestimate the bandwidth | | The Silk Road Group, Ltd. | of an in-line skater with tapes | | | hurtling down your favorite trail."| | http://www.silkroad.com/ | -Tanenbaum & Bass | | | | +--------------------------------------------------------------------------+
Just because someone has 30% of the internet, they still have an interest in connecting their 30% of the net to .1% of the net, no?
Because they can already hit that .1% of the 'net already without peering with that smaller provider through that smaller provider's transit provider and peering: 1) Enables them to screw you up a little more than they otherwise could routing-wise 2) Provides them with free 'transit' across your backbone (even if it's to/from your customers) if the peer isn't multiple-exchange-point connected to some of the exchanges you are. 3) Has those political 'what is a peer' problems also.
-alan
Avi
Let's look at a hypothetical situation. ISP1 peers at MAE-E and buys transit from MCI there. Now they ask Sprint to peer with them. Let's look at how they reach sites on the west coast. Without peering, MCI gives the packets to Sprint at MAE-East and Sprint returns packets to MCI at some west coast nap. That is, MCI and Sprint share the coast-to-coast traffic. With peering, Sprint must take the packets all the to MAE-E as it has a shorter AS path. All the coast-to-coast cost is borne by Sprint. Do you get that? Now do you understand the 3 NAP rule? David Schwartz On Thu, 9 May 1996, Alan Hannan wrote:
So, comes my curiosity, and my puzzling thoughts about the current state of the net. Why is it not in my best interest to talk to NSPX at a meet point? Why, when it is in MY customer's best interest to talk to EVERYONE, would I not converse, and share knowledge and invitations about my customer base?
Let's look at a hypothetical situation. ISP1 peers at MAE-E and buys transit from MCI there. Now they ask Sprint to peer with them. Let's look at how they reach sites on the west coast.
This is a violation of the rules of Mae-East: it is not to be used for customer connections. ISP1 must have a private connection to MCI. Erik
Without peering, MCI gives the packets to Sprint at MAE-East and Sprint returns packets to MCI at some west coast nap. That is, MCI and Sprint share the coast-to-coast traffic.
With peering, Sprint must take the packets all the to MAE-E as it has a shorter AS path. All the coast-to-coast cost is borne by Sprint.
Do you get that? Now do you understand the 3 NAP rule?
David Schwartz
On Thu, 9 May 1996, Alan Hannan wrote:
So, comes my curiosity, and my puzzling thoughts about the current state of the net. Why is it not in my best interest to talk to NSPX at a meet point? Why, when it is in MY customer's best interest to talk to EVERYONE, would I not converse, and share knowledge and invitations about my customer base?
On Fri, 10 May 1996, Erik Sherk wrote:
Let's look at a hypothetical situation. ISP1 peers at MAE-E and buys transit from MCI there. Now they ask Sprint to peer with them. Let's look at how they reach sites on the west coast.
This is a violation of the rules of Mae-East: it is not to be used for customer connections. ISP1 must have a private connection to MCI.
True. Serves me right for composing a quick example. Let's change the example to say they get the transit through a private T3. DS . So ISP1 peers at MAE-East and has a private serial circuit to MCI (or someone elses) router at the MAE (not using MAE infrastructure). Same result. And same result even if ISP1 peers at the MAE and sends all its traffic to any MCI site 'near' (in network terms) MAE-East, ust as long as MCI gives the traffic to Sprint there. Alex Bligh Xara Networks
In message <199605100309.WAA25431@westie.gi.net>, Alan Hannan writes:
So, WHY would an NSP enter into a peering agreement with another person? Why, to profit from the one side of the connection, to enable an entity [labeled A] to talk with some other entity [labeled B]. In most cases, NSP1 had as customer A, and NSP2 had as customer B, and obv. it was in their best interest to meet somewhere to talk to each other. NSP1 added value to A, by providing a path to B. NSP2 added value to customer B by providing a path to A.
Maybe because their routers fall over beyond some number of direct peers and they have a problem with the idea of going through the RA's router servers to avoid that problem (whether the problem is real or imagined, present company included - not sure if a smiley or a frown is needed here). This results in a need/desire to prioritize. Let me turn the question around a bit. Suppose N small providers peer at M places and together represent p% of the Internet. If they simply appear at one NAP and don't contract for transit, they may reach 100%-(%p/(M-1)) of the Internet. Is that something to encourage? If they must contract for transit and as a result reach all major interconnects, they get 100% (possibly minus a small epsilon for other reasons). They then don't need to be at any of the NAPs. Are some providers trying to show up at one NAP only with the aim of not contracting for transit through anyone even though they can't really reach others like themselves at a different interconnect? Curtis ps- this is a tangent but - I think ANS has a problem with using the RS since at one point we announced everything to the RS and started giving transit connectivity to people that were not transit customers of ANS. The technical barrier to fixing this at that time was an inability to express outbound policy adequately (no support for as-paths). This technical barrier is almost gone except we have code that uses as-macros and we'd have to exapnd the as-macros and expresses this policy in our aut-num. So the ball is in our court (along with too many others). The other issue is we have no management access to the RS and so if any of this isn't configured and/or working right on the RS we can't easily tell if it is. The issue in general for most providers is they want direct control over their peerings and import and export policy. pps- I suppose the other solution is to get routers that can handle the peerings directly. :-)
participants (6)
-
Alan Hannan -
Avi Freedman -
Curtis Villamizar -
David Schwartz aka Joel Katz -
Erik Sherk -
Tim Bass