Talk:Internet exchange point

Middle East IXPs

Please read the following notes, before editing the Middle East section again:

- There was briefly a government-initiated IXP in Cairo, called the CR-IX, in 2002. To the best of my knowlege, it was wholly defunct by 2004. It was succeeded by a much quieter private-sector IXP, which currently has three participants, on 81.21.96.0/25.

- There is an IXP planned for Bahrain, called Gulf-IX, in the still-under-construction GatewayGulf datacenter. With luck, it will be up sometime in mid-2007. There is no other IXP in Bahrain. The "Bahrain Internet Exchange" is a small government-owned transit provider.

- EMIX, the Emirates Internet Exchange, likewise, is a transit provider, not an IXP. They'll tell you so themselves.

There's also the small IXP in Tel Aviv, which hasn't changed much in a long time. So: please only edit this section again if you have some good news for us. Bill Woodcock 22:17, 2 January 2007 (UTC)

Chinese IXPs

My understanding was that the three IXPs in China were state-controlled and operated in Beijing, Guangzhou, and Shanghai. See here.140.247.147.88 01:07, 20 February 2007 (UTC)

New York

What about that huge exchange point in New York City? --Abdull 16:31, 1 August 2005 (UTC)

afaik the NYIIIX (New York Internet Exchange) is rather a small one. Have a look at the german page, there you see the traffic of Internet Exchange Points. -- Alvo 10:19, 24 September 2005 (UTC)

The NYIIX is quite large, for a U.S. exchange... But more to the point, they're in 111 8th, along with PAIX and Equinix, and the crossconnects in NYCConnect, and the new NYCJ 10gbe exchange. New York is slowly taking the place of D.C. as the east coast regional exchange. --Bill

Jecowa is suggesting merging National_Inter_Connection_Exchange and Indonesia_Open_Internet_Exchange_Point into the hierarchy of this page. I'd support that, with the caveat that it would be nice if someone could provide independent verification of the existence of either of these IXPs. Looking glass? --Bill

Wrong content in wrong page

I think we should have an article that explains what an exchange point is. Plain and simple. All other internet exchange points and their detailed infomation should be in different articles. Ken

Also, there are a several good listings of IXPs in the Internet: Packet Clearing House, Open Directory Project, Exchange Point Repository, BGP Forum, EuroGIX. Probably we dont need to maintain a list of IXPs in a wikipedia. In the german wikipedia, there have been artifcles about IXPs removed because they have been considered as not relevant in a encyclopedia. Regards -- Alvo 18:56, 11 December 2006 (UTC)

I'm of mixed minds about this... I run Packet Clearing House, and we've maintained the large directory of IXPs since about 1994 (well, it was only a couple, then, but it was equally complete), and put quite a bit of staff time into it. At one level, it's redundant for us to also be trying to copy updates out to Wikipedia. On the other hand, once or twice there have been people who've thought to post an update of some sort to Wikipedia before we've discovered it otherwise. In general, though, I agree that differentiating definitions and directories is a good thing. Bill Woodcock 20:42, 15 December 2006 (UTC)

Bulgaria

I believe the two exchange points mentioned are defunct. Peering is currently implemented via private gigabit ISP to ISP connections using a MAN.

Neutral Access Point

I understand that the term NAP is still used in Europe, eg http://it.wikipedia.org/wiki/Internet_Exchange_Point and Google shows many hits. I think NAP should be included under IXP. Regards John John a s 10:14, 27 August 2007 (UTC)

OK, I think there was already a discussion about that in: Talk:Network_access_point I think if there are no vote against it, I can try to merge the two articles and redirect NAP to IXP. Any comments? --Never stop exploring (talk) 09:28, 18 April 2015 (UTC) PS: Let me say hi to User:Bwoodcock :)

I don't hear _Neutral_ Access Point used much at all, honestly. When people say "NAP," they mean "Network Access Point," in the historical or Latin-American senses, generally. And I assume this is Marco? Anyway, yes, I'd support a merge. Bill Woodcock (talk) 01:30, 19 April 2015 (UTC)

Pakistan

I've removed PIE from the list of IXPs. It does not appear to be an IX, just a telco-owned transit provider network. See e.g. [1]. Compare Bahrain Internet Exchange, EMIX, and so forth. Will-h 15:27, 23 September 2007 (UTC)

Possible merge with the NAP article

On 6 June 2012 user Ál added a merge template suggesting that the NAP and IXP articles be merged. Other than the merge templates, I don't see any further discussion about or rational for the merge.

For myself, I don't think a merge is a good idea. NAPs are largely historical now, while IXPs are current. It seems like adding historical information about NAPs to the IXP article would be just adding unnecessary clutter. --Jeff Ogden (W163) (talk) 14:52, 2 July 2012 (UTC)

I removed the merge template from the article. If other editors still feel that a marge is a good idea, we can continue the discussion at Talk:Network access point#Possible merge with IXP article. --Jeff Ogden (W163) (talk) 18:25, 9 September 2012 (UTC)

It's still a good idea. I'm giving it a try as a new "History" section here. If there's general agreement on both sides, then we'd turn "Network Access Point" and "NAP" into redirects to the IXP article, after deleting the old Network Access Point article. NAPs are entirely deprecated, and are exclusively a historical subset of IXPs, so it seems appropriate, and if they're not a single article, there will be people who will find their way to the NAP article and not follow to the IXP article, and remain confused re context and scope. Thus the rationale for the merge. Bill Woodcock (talk) 13:13, 11 August 2021 (UTC)

Inapplicable / misleading diagram

Is anyone really wedded to the diagram on the bottom right, https://en.wikipedia.org/wiki/File:Internet_Connectivity_Distribution_%26_Core.svg ? It's misleading in the sense that it features "tier" terminology heavily, as though that meant something specific or useful, and it implies that "tier 1" and "tier 3" ISPs don't peer at IXPs. I'm pretty sure this does more harm than good. Bill Woodcock (talk) 22:09, 30 December 2013 (UTC)

I'm not wedded to this particular diagram or its terminology. I do think that some sort of diagram is useful in this article. --Jeff Ogden (W163) (talk) 23:56, 30 December 2013 (UTC)

Agree with Jeff Ogden (W163), this sort of diagram is *not* useful in this article. Wonder if there is any good source or a better diagram from someone else, did you had something in mind Bill Woodcock? --Never stop exploring (talk) 10:35, 18 April 2015 (UTC)

Just replaced the offending diagram. If anyone disagrees, please pipe up, or revert, or whatever. Thanks. Bill Woodcock (talk) 02:03, 19 April 2015 (UTC)

Misleading first sentence

The first sentence of this article reads "Internet exchange points are the locations where Internet bandwidth is produced, analogous to factories for physical goods". In my opinion this is an actively misleading description, and I was alerted to it by someone who is learning about internet technology and was quite confused by it.

Bandwidth is not "produced" in any meaningful technical sense. Internet bandwidth is a property of connectivity; it is not produced anywhere. If you visit the website of any IXP, you will not anywhere see any mention of them "producing" bandwidth. The Internet Exchange Federation's definition of an IXP does not mention this either.

The cited paper is making an economic analogy with bandwidth as a physical good, and while it may be useful for the purposes of this paper (which appears to be aimed at policymakers), I do not believe it belongs in an encyclopaedic description of what an IXP is. I will also note that this sentence appears to have been added by one of the authors of the paper it cites.

I would recommend that this sentence is removed. The following sentence, while not perfect, is a much more accurate description of what an IXP is. I attempted to make this change but it was reverted. - Russ (talk) 17:25, 5 October 2021 (UTC)

Agreed, this is a really confusing definition. Bandwidth isn't "produced", merely traffic exchanged. A more useful definition might be the IX-F definition of an IXP at http://www.ix-f.net/ixp-definition.html - which is basically what Russ wrote. Perhaps it would be better to cite this definition in the article? Will-h (talk) 17:32, 5 October 2021 (UTC)

I absolutely disagree. You're suggesting that we remove the context and leave only a bare description. The Wikipedia article for "money" doesn't say that it's rectangular pieces of paper. The Wikipedia article for "cars" doesn't say that they're metal boxes. If Wikipedia were written for raccoons or something, those descriptions might be sufficient, but Wikipedia is an encyclopedia, written for humans. More than half of all humans consume Internet bandwidth. In the context of humans, the interesting thing about IXPs is that they're where the bandwidth comes from, not that they're rooms with cables going in and out. Keep the first sentence as-is. It's concise and accurate. BurritoTunnelMaintenance (talk) 13:08, 6 October 2021 (UTC)

BurritoTunnelMaintenance is correct. The current leading sentence makes sense, and explains what IXPs do and why they exist. Wikipedia is an encyclopedia of articles, not a catalog of descriptions. Reducing this to a description would make the article pointless. RLMcGinley (talk) 14:38, 6 October 2021 (UTC)

Please see Bandwidth_(computing): "maximum rate of data transfer across a given path" -- this is not something which can be "consumed", any more than turning on the tap in your house to have a drink consumes water pipe. Users receive and consume internet traffic in the form of packets. Moreover these are *not* generated in an IXP; typically these are generated in a Content_delivery_network which usually perform interconnection using IXPs, as well as interconnecting via other methods. Characterising IXPs as a "source of traffic" in this way is entirely misleading. Will-h (talk) 13:23, 6 October 2021 (UTC)

I'm reasonably familiar with both the spectrum and Internet-economic uses of the word "bandwidth." It's a word which is confusingly overloaded, and that's unfortunate. As a thought experiment, let's replace it with "foo" for a moment. "Foo" is what you're paying for, when you buy service from an ISP. "Foo" is what they're selling you when they take your money. When you don't get "foo" from your ISP, you become dissatisfied and switch to a different ISP. Now, where does the ISP get the "foo" that they're selling you? Do they have a factory out back, where they produce it? Do they harvest it from a field? Do they use big nets to collect it? Do they buy it from someone else, and if so, where does that person get it? Bill Woodcock (talk) 14:25, 6 October 2021 (UTC)

The first sentence is as consice and accurate as "Traffic is produced by intersections". While I agree that we should write a first sentence in a way that it's easily understandable, we should stick with what's correct. 1183519e (talk) 13:49, 6 October 2021 (UTC)

It is simply incorrect to say "bandwidth is produced". It's not a commodity, it's a property of a connection. The cited paper is stating this as part of a (somewhat tortured) economic analogy. This analogy might be appropriate further down the article, but placing this in the first sentence is misleading. I am more than happy to try and improve the first paragraph of this article to make it more accessible to the layperson, but I can't do so without removing this statement first. Russ (talk) 14:15, 6 October 2021 (UTC)

Hi. It appears to be my paper that's being cited, but this is a relatively common understanding among people who study telecommunications economics, albeit that's not a widely-published field. I've never heard this understanding called controversial before. What, exactly, do you believe to be incorrect about it? Do you believe that Internet bandwidth is a naturally-occurring phenomena? ISPs just gather it up somewhere, when nobody's looking, and sell it to their customers? If this isn't what IXPs do, what do you believe the economic function of an IXP is? And where do you think bandwidth comes from, specifically? Bill Woodcock (talk) 14:21, 6 October 2021 (UTC)

I do understand the economic perspective (although another quibble: given the increasing prevalence of edge caches, I would be surprised if IXPs are responsible for even a majority of end-user bandwidth these days, while the article implies they are responsible for all of it), but I am quite sure that this use of terminology would be surprising to the vast majority of people who work in the networking field. Someone who is learning about networking concepts is going to find this statement confusing - in fact, I only found this page because someone else was confused about it. The word "production" here is the economic jargon, but this is not an article about economics. I'm not averse to this being used further down the article, but I don't think it belongs in the introduction. Russ (talk) 14:39, 6 October 2021 (UTC)

You really believe that the word "production" is impenetrable jargon which a Wikipedia reader cannot be expected to understand? In addition to the Production (economics) page, there are also disambiguation links to "Production: the act of Manufacturing goods" and "Production, in the outline of industrial organization, the act of making products (goods and services)" and links to Means of production and Product (business), all of which seem to me to be clear enough, and all of which use the word "production" in the same sense that I and others do when we discuss the production of Internet bandwidth. Indeed, the overloaded term "bandwidth" is almost always the source of confusion, in my experience... lay people understand it to mean what I and economists understand it to mean, while engineers mostly seem to default to the unrelated use of the word with respect to spectrum, or back themselves into unnecessary extremist positions on the red-herring of where on the goods-services continuum Internet bandwidth lies. So, perhaps you're correct that, while this usage is intelligible to the vast majority of people, the subset who are familiar with RF engineering but not Internet cost accounting may find it confusing. I would argue that Wikipedia's primary purpose is to serve the general public, not the small population of people who misapply a term from an unrelated field. As for the assertion that an unknown but larger portion of bandwidth production is from on-net sources rather than interconnection, I would encourage you to find numbers to support that... I've been looking for more than twenty years, and have failed to find any, whereas many mass-traceroute studies show otherwise. But, anyway, your assertion gets deep into WP:NOR. If you can find supporting data and publish it, you will indeed be lauded. And, undoubtedly, richly rewarded by AT&T. It is a fact that a portion of bandwidth is produced by on-net sources, and what that portion is varies from party to party; it's been a long time indeed since I've seen anyone try to support a figure greater than 1%, though, based on either actual measurements or customer market-share. Even a fraction of 1%, though, would make it correct to say that not all Internet bandwidth is produced in IXPs. So, I would support additional text clarifying that. I'm not sure that it belongs in the intro paragraph, but it's certainly a detail worth mentioning, for those interested in delving further. Bill Woodcock (talk) 21:46, 6 October 2021 (UTC)

Where to start… okay, first:

• ISPs just gather it up somewhere, when nobody's looking, and sell it to their customers

— please do not engage in Reductio ad absurdum as an argumentative technique. This is not conducive to a positive, respectful exchange.

It's an honest question. If you have an answer for it, please provide it. That's how discourse and the general understanding in the field moves forward. Bill Woodcock (talk) 15:52, 6 October 2021 (UTC)

It would've been an honest question, had you left out the absurd, extraneous ISPs just gather it up somewhere, when nobody's looking, and sell it to their customers. I'm sorry but I am not interested in spending my time on a discussion with you on this topic, particularly considering you have a personal commercial interest in the very subject of this Wikipedia article due to your position as executive director at PCH. I would kindly ask you to recuse yourself from this discussion as well as any future editing of this article.

Per the rules of Wikipedia:Conflict_of_interest, I'm (pre-)notifying you of your involvement in a conflict of interest discussion. I haven't made the final step of throwing this onto the CoI notice board yet, wanted to give you a chance to comment first. --Eqvinox (talk) 16:52, 6 October 2021 (UTC)

Comment on what, exactly? I'm very curious to hear what this "commercial interest" you think I have is. Please elaborate. Bill Woodcock (talk) 16:57, 6 October 2021 (UTC)

As executive director at PCH, an increase in IXP operations is beneficial to your business. This directly aligns with the overstated importance of IXPs that is currently expressed by the introduction to this article. --Eqvinox (talk) 17:06, 6 October 2021 (UTC)

Ok, there are several assertions there; can you connect them to form an argument? I am indeed executive director of PCH, but that appears to be unconnected with that which follows. I do indeed have business interests. How would an "increase in IXP operations" be beneficial to any of them? Please suggest some specific transaction which you believe might exist. And, tangentially, I'm curious what you mean by an "increase in IXP operations." We've gone from zero IXPs to roughly 800 while I've been doing the various things I do so, depending what you mean by "increase in IXP operations," it wouldn't appear to be hypothetical nor can I see how any rate of change in that increase would hinge on a few words of a Wikipedia article. Nor, to get back to the meat of the matter, can I see a commercial interest, but as I said, I'm very curious to hear what you think exists. Bill Woodcock (talk) 19:39, 6 October 2021 (UTC)

Quoting pch.net: The task that PCH was first created to address is the building and support of Internet Exchange Points. IXPs are the places where Internet Service Providers interconnect their networks, to exchange traffic between their customers and create Internet bandwidth. PCH assists in the construction of IXPs, and provides equipment, training, and ongoing support to IXPs and their operators.

I'm arguing that this is a rather obvious conflict of interest; in a hypothetical world without IXPs, PCH would not exist at all. --Eqvinox (talk) 20:34, 6 October 2021 (UTC)

Forgive me, but if it were "obvious," I suspect you'd be able to actually state what it is. Indeed, in a hypothetical world without IXPs, a portion of what PCH does would not be necessary, we might have a different name, different board composition, etc. But we're in an actual world, and you're making an actual assertion of conflict of interest, so it seems as though the next step would be to define what that conflict might be, rather than hypothesizing alternate-world scenarios, no? Wikipedia does in fact have CoI policies, WP:COI, and definitions of what constitute such conflicts, WP:EXTERNALREL. Unlike many of the editors of the article, and indeed some of the contributors to this very conversation, I am not, nor ever have been, an employee of, nor received compensation from an Internet exchange point, nor do I, or any organization I receive compensation from or have ownership in, receive money or value for the provision of IXP services. Nor does anyone I'm related to by birth or marriage. Further, Wikipedia policy goes on to say: "Subject-matter experts are welcome on Wikipedia within their areas of expertise, subject to the guidance on financial conflict of interest and on citing your work." Both of which I have been scrupulous to do, for as long as those policies have existed. Many others have cited my work in the article, and on the occasions when I was able to clean something up and was able to cite longer-form work published elsewhere with supporting detail, I've done so, whether I was the other or others were the authors. If you were to, as you suggest, strip the article of contributions by those who, unlike me, do have a financial interest in IXPs, you would be truly gutting the article. In any case, I am honestly intrigued by the notion that there's a conflict here, so, again, please elaborate: what is the "commerce" or "business" to which you refer? I'll happily continue the discussion until it's resolved. Bill Woodcock (talk) 21:46, 6 October 2021 (UTC)

• a relatively common understanding among people who study telecommunications economics

— this is completely meaningless here, Wikipedia is not written for people who study telecomms economics, and the terms used in some specific field can be massively misleading to the general populace.

Are you arguing that the concept of or term "production" is misleading, or that the concept of or term "bandwidth" is misleading, or something else? I would argue that both the concept and terms "production" are generally well-understood by the public, and that the term "Internet bandwidth" is generally understood by the public to be the thing that they're buying. If you were to argue that it should properly be called "transit," I would agree, under some circumstances, but that is not a term that I believe is commonly understood by the general public. Bill Woodcock (talk) 15:52, 6 October 2021 (UTC)

• where do you think bandwidth comes from, specifically?

→ Usable internet bandwidth is a product of infrastructure being built.

Usable televisions are a product of television factories being built. The latter half of the sentence analogizes to factories. I'm not wild about trying to understand things by analogy, because it never gets you very far and, yes, can be misleading. But talking about something coming into being "as a product" of the means of production being built seems unnecessarily obtuse to me. You can build a factory and never put it into production, and nothing will be produced. The product does not come into being without the factory being exercised. Thus I would say that the infrastructure being built is a necessary precondition, but not sufficient. Just as the ISPs connecting to the IX is a necessary precondition, but not sufficient. Bill Woodcock (talk) 15:52, 6 October 2021 (UTC)

→ Installed cabling or RF spectrum has a limited amount of maximum extractable bandwidth based on physical limitations. Telecomms hardware makes use of some of this bandwidth and builds interconnectivity on top.

You appear to be conflating the spectrum and economic senses of "bandwidth," which greatly confuses the conversation. Do you have a point to make about spectrum, because if so, it's not apparent what the point is, or how it relates to IXPs. Bill Woodcock (talk) 15:52, 6 October 2021 (UTC)

→ Actual internet bandwidth you can sell to a consumer is only produced when one or more paths from the consumer to the wide majority of internet services come about as a result of both transport capacity (cabling) and interconnect capacity (routers) combining to provide this service. An IXP facilitates one such interconnect. It is by no means "a" or "the" "source" of bandwidth. In general, an IXP needs to be understood as a shortcut in facilitating a trade; since no single internet provider "provider-s" the entire internet, making sure paths (to "produce bandwidth") are available always requires cooperation between multiple ISPs. IXPs are one such avenue of crossconnection between ISPs, and generally a cheap and efficient one, contrasted against expensive agreements with larger ISPs to purchase connectivity from them.

That's conflating peering with transit, and the latter is certainly a non-sequitur in this context. Peering and transit have their own Wikipedia articles, this one is about IXPs, which are the loci at which peering occurs. Which would be a reasonable definition if it were intelligible to the public, but I recognize that it's not. I suggest that we leave trying to redefine transit and peering to another argument on another day. Bill Woodcock (talk) 15:52, 6 October 2021 (UTC)

→ Connectivity can be - and is frequently - built entirely without any IXPs, particularly by private crossconnects between larger networks.

...which occur where, exactly? If you get rid of all of those interconnections, what's left, that anyone would pay for? The largest market-share any ISP has ever had was UUNet, in the late 1990s, and even then, on-net traffic was single-digit percentages. Bill Woodcock (talk) 15:52, 6 October 2021 (UTC)

• If this isn't what IXPs do, what do you believe the economic function of an IXP is?

→ and, lastly, this is not what this Wikipedia article needs to answer, and likely even irrelevant to the article to begin with. The question to be answered here is „what is an IXP?″ in a general, layman, encyclopædic sense.

"A place with cables." But I think you're putting the cart before the horse here. One could categorize the people of the world into two groups: a group which cares about IXPs for their own sake as a place with cables; and a group which care about IXPs for their functional value as the places where their bandwidth comes from. I suspect that the former group has, perhaps, a few dozen people in it, while the latter has, at last count, about 4.6 billion. I suggest that the value of a Wikipedia article comes from its utility to the latter group, rather than how satisfying it may be to the former. Bill Woodcock (talk) 15:52, 6 October 2021 (UTC)

I may take a stab at formulating something less… weird… in a bit. --Eqvinox (talk) 14:59, 6 October 2021 (UTC)

If bandwidth is "not produced anywhere" and it's a "property of connectivity," why do you think people bother to build IXPs, and what do you think ISPs do for a living? EVhotrodder (talk) 14:33, 6 October 2021 (UTC)