[NNagain] The Whys of the Wichita IXP Project

[rjmcmahon]

Boston University spent $305M on this and it doesn't have an IXP.

https://www.bu.edu/articles/2022/center-for-computing-and-data-sciences-photo-essay/

It's like building a magnificent train station w/o any tracks to/fro the 
station.

Bob
> On Wed, Feb 21, 2024 at 8:02 PM Brent Legg via Nnagain
> <nnagain at lists.bufferbloat.net> wrote:
>> 
>> First, let me offer a public THANK YOU to Dave Taht for reaching out 
>> to us about the specifics of our Wichita IXP project, and for inviting 
>> me to join this group.  It’s been disheartening to see folks talk 
>> about us & the project on public forums like LinkedIn without first 
>> engaging us in conversation to learn the specifics of what we’re 
>> actually doing.  I’d like to think that those who have been 
>> disparaging have only done so because they don’t understand what we’re 
>> trying to achieve.
> 
> I was wildly enthusiastic to see what you were proposing appear in the
> press. It was a breath of potentially fresh air in an otherwise
> depressing post RDOF, post BEAD environment where it seemed like the
> only metrics were speedtests and passings.
> 
> I try very hard to get people of wildly disparate backgrounds to
> converse, and escape the bubbles they are in. I have tried to gather
> together on this old-fashioned email *discussion* list both
> technologists and policy-makers to clear the air in ways that cannot
> be encapsulated in 240 characters. These two groups (a lot of old
> internet experts here) have not been communicating very well of late,
> ironically, over the best communication medium ever invented.
> 
> It is sad that email lists have so been in decline the past 20+ years,
> overwhelmed by marketing and spam, as an email address is the only
> universal identifier we have for so many other transactions. The
> advantage of a discussion list, over all the faddy technologies, are:
> you retain a copy of what you said, everyone else does also, and the
> internet at least used to make it searchable into the far future.
> Remembering that I had a dispute or discussion with @randomperson and
> finding them again via the technology-of-the-day (g+ anyone?, slack?
> disquis? hackernews?) is really hard otherwise, and I do hope that
> email makes a comeback.
> 
> But someones need to start maintaining it better.
> 
>> 
>> 
>> To begin, I think there is confusion in the terminology being used.  
>> When we say “IXP,” we mean the facility (building, venue) where 
>> interconnection & peering occurs.  The “IX” is the ethernet switch in 
>> the building.  When someone says an IXP can be built for $8k, that’s 
>> apples-to-oranges with what we’re doing.  Yes, a switch can be 
>> procured for $8k.  But where does it go?  What if there is no safe, 
>> secure, neutral place for it to go?  Then such a place must be built.  
>> That’s what we’re building in Wichita.
> 
> To not annoy us old farts, clarifying that you mean a carrier neutral
> facility or datacenter with an IXP would go a long way. :)
> 
> Too many in the past built gold-plated IXPs, ending up with an
> appalling cost model that attracts nobody. This total plan,
> at this cost, is a *very good* one, and my hope would be, commoditized
> and widely replicated to even more than the 120 locations you project
> - but my hope is that the IXP component will mirror the successful IXP
> models already existing in the USA.
> 
> The costs of interconnecting networks have fallen dramatically, and
> can fall further.
> 
>> 
>> 
>> Saying an IXP can be built for $8k is enormously confusing to many 
>> policymakers who do not understand the issue or how interconnection & 
>> peering actually work, yet have enormous power to set policy and spend 
>> money that will affect the future of the Internet for generations.
> 
> Operational expense needs to be discussed. The underlying technologies
> used to "make it happen", need to be selected. It is amazing what a
> modern cheap 100GB 32 port switch can do. IPv6 is mandatory nowadays
> while still finding a way to carry what little remains of IPv4 space
> efficiently is needed. It would help if there was a local mirror of
> one or more of the root DNS servers. Some really tough design choices
> regarding what forms of active ethernet fiber vs a vs gpon need to be
> made. And so on. Who makes those decisions?
> 
>> 
>> 
>> 
>> We began this whole initiative by asking a series of questions to help 
>> us arrive at our model for IXP (building) proliferation.  I’ll use 
>> Wichita as the context for these questions, but these could just as 
>> easily apply to any other similar city that is home to a large public 
>> research university:
> 
> Thank you for sharing this last criterion. I had done a similar (much
> briefer) study targetting latency and resilience primarily, and what
> it would cost to do more "rural IXPs" - call them RXPs - every 50
> miles or so - on the cheap as an outgrowth of BEAD. But that would be
> a subject for another thread.
> 
> But I did not limit it to "research" universities, but to areas that
> had universities. Certainly there is high demand for sexy AI-related
> things, but the nuts and bolts of how to design and build networks, is
> lacking.
> 
> I regard network design and operations to be a branch of civil
> engineering nowadays, and most operations people are quite leery of
> letting grad students loose with operational networks. I would love to
> see more universities actually teaching the skills to be a decent
> sysadmin (or SRE), because basic knowledge of packets, routing, tcp,
> bgp, resiliency, and so on is in the decline. Being a BOFH requires
> far more skills than a electrician and is actually comparable in
> skills and stress to being a doctor. (SREs get paid pretty well, but
> most fall into the profession rather than being directly trained on
> it)
> 
> Instead, I have been coping (as part of bead), at 6 week educational
> programs intended to train people how to splice fiber.
> 
> So I would broaden your targets to places that also intend to teach
> people how to design and maintain civil infrastructure, and plan ahead
> for disaster recovery. This includes connecting up governments and
> emergency services. Reusing old postal buildings is an option, as are
> other lower grades of schools.
> 
> I would love to see curricula for the next generation of BOFHs that
> included formerly basic things like how to decode a packet capture and
> teachings from TCP/ip volume 3, illustrated, and everything
> in-between.
> 
> Obligatory xkcd: https://xkcd.com/705/
> 
>> 
>> 
>> 
>> Should Wichita, with a regional metro population of 600k+, be 
>> literally dependent, from an interconnection standpoint, on Kansas 
>> City and Denver forever?  No.
>> Okay, then what type of facility does Wichita need?  Ideally, 
>> something that can meet current needs and scale to meet future needs.
>> What are the attributes of such a facility?
>> 
>> Does it need to be carrier-neutral?  Yes.
>> Does it need to be secure?  Yes.
>> Does it need to provide a level-playing field for networks of all 
>> types?  Yes.
>> Does it need to be able to convey rights to, and protect the rights 
>> of, its tenants?  Yes.
>> Does it need to be a facility that networks can rely on to remain “up” 
>> in the wake of adverse events?  Yes.
>> 
>> Resilient from power outages?  Yes.
>> Resilient from cooling equipment failures? Yes.
>> Resistant to wind damage?  Yes.
>> Resistant to vandalism or ballistics damage?  Yes.
>> 
>> Does it need to be financially sustainable?  Yes.
> 
> So that is the good question. How do you do opex?
> 
>> Is “best effort” good enough?  No.
> 
> Redundancy helps.
> 
>> Then does it need to be professionally managed?  Yes.
> 
> Where will they come from? What software do they have to manage the
> facility? Who writes the software?
> 
>> 
>> Is there an existing facility in Wichita that can meet those needs?  
>> No.
> 
> In general I use latency as a proxy for where interconnects should go.
> Historically this has been about 500 miles. I thought it was
> interesting to explore what (as part of Biden´s ev charger program)
> what it would take to have an old fashioned IXP ever 50 miles. Turns
> out that is pretty close 8k in gear + a lot of fiber.
> 
>> So one must be built?  Yes.
>> Where should it be built?  Where a concentration of eyeball traffic 
>> already exists that can grow a peering ecosystem faster than it might 
>> otherwise, and that is also proximate to existing fiber plant, and 
>> where diverse manholes can be placed on the edge of public 
>> right-of-way.
>> 
>> 
>> 
>> In the case of Wichita, that’s at Wichita State University.
> 
> Do they teach how to run a network?
> 
>> 
>> 
>> 
>> Creating a secure, neutral, resilient interconnection facility with 
>> proper cooling, power systems, lockable cabinet space, diverse 
>> manholes and POE isn’t cheap.  The whole project is actually more than 
>> the $5M grant we received.  We’re putting in over $800k in cash, plus 
>> additional in-kind match.
>> 
>> 
>> 
>> We’ve done the data analyses necessary to determine which communities 
>> need such facilities, and that’s how we came up with our list of 125 
>> target communities.  Most of them are home to public research 
>> universities, but have no IXP or IX.  Not all of those communities are 
>> equal in terms of priority, but all of them have a need, and we’re 
>> actively seeking pathways to scale that preserve our core principles 
>> and avoid the need for grants.  But that’s a big challenge.
>> 
>> 
>> 
>> I really appreciate the opportunity to provide clarity on the project 
>> and I’m happy to answer your questions.  Surely we agree on much more 
>> than we disagree.
>> 
>> 
>> 
>> --Brent Legg, Connected Nation
>> 
>> _______________________________________________
>> Nnagain mailing list
>> Nnagain at lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/nnagain