>>> On 6/15/16 05:37, Mike Hammett wrote: >>> A non-profit donation-based IX that doesn't produce results >>> could be screwing its "customers" over more than a MRC-based >>> for-profit IX that does produce. >> >> On 15.06.2016 21:14, Seth Mattinen wrote: >> An IX just needs to "produce" a layer 2 peering fabric. That's not a >> tall order to get results from. Anything beyond that is extra fluff. >> Some people want to pay more for the fluff, some don't. > > On Jun 15, 2016, at 6:36 PM, Arnold Nipper <[email protected]> wrote: > This is a *common* misunderstanding. > The by far easiest part of running a successful IXP is the technical part. > The more challenging is to build a community around it. And that's > purely non technical and involves a lot of *social* networking and > bringing people together.
There’s a difference between the cost and the product. As regards the cost,
Arnold is exactly right. Across the many hundreds of exchanges that we’ve
worked with over the past 22 years, our observation has been that, at a rough
average, most IXPs spend 45% of their first-year effort on location selection,
45% on governance definition and establishment, and 10% on technical decisions
and implementation. But the total effort and the governance portion both
increase drastically for those that choose to handle money; at a very, very
rough average, about four-fold. In subsequent years, location selection
generally drops away to near zero, except in cases like the JINX, and technical
work dips for the first couple of years, and then spikes once every three years
or so as switches are replaced and new configs are needed. Many exchanges have
an annual in-person meeting where elections are conducted and policy changes
ratified, so that typically becomes the largest ongoing expense, as Arnold
implies.
As regards the product, no, Seth, the layer 2 peering fabric is merely a
necessary precondition for producing bandwidth. The actual bandwidth
production has other preconditions as well: peers physically connected to the
peering switch fabric, BGP sessions established between the peers, routes
advertised across those sessions, a reasonable matching of potential traffic
sources and sinks available through those routes, and a set of customer
behaviors that prefer those source/sink matchings. Only then does an IXP
produce bandwidth. So, the role of a salesperson or advocate or evangelist or
tout can be a net beneficial one, if they do a good job of recruiting
participants, making sure they follow through with peering, and encouraging the
preference of locally-available content. WAIX was among the first IXPs to do
this well, in my opinion.
-Bill
signature.asc
Description: Message signed with OpenPGP using GPGMail
