[*] CIRA statement on IXPs

[John Lange]

I'll resist rehashing all my previous poo-pooing of this idea, but it
comes down to one simple thing. Money. If there was any business case
for this, it would have happened already.

The PDF document, in my opinion also makes some (unintentionally)
misleading statements.

For one, it says there are only 2 IXPs in Canada. While it's true
there are only 2 independent IXPs, there are many many more places
where peering takes place between 2 carriers. And this fact is
fundamentally is why there aren't more IXPs (they aren't needed).

"Bell Canada does not peer in Canada"

I'm not sure what data they are relying on for this, but the BELL-GT
AS (http://fixedorbit.com/AS/6/AS6539.htm) shows peers with Shaw,
Rogers, Allstream, even MBHydro.

Never the less, the one interesting point raised by the document is
security. Many Canadian companies are prohibited from buying cloud
services from providers outside of Canada due to privacy laws.
However, even if the provider is in Canada, according to the document
there is still about a 35% chance your data goes through the US. I'm
wondering if technically this is still a violation and if so, might
this force Bell (who is a big offender according to the document) to
peer in Canada since their clients would be forced to change
providers?

"An IXP ordinarily offers the greatest benefit to a network if the IXP
and network are proximate ... (in the same building)"

This is why TORIX exists. Everyone was already in 151 Front so they
asked building management if they could setup a meetme room which 151
Front (according to the PDF) provides for free. That situation is
more-or-less unique in Toronto and is why economically its challenging
any place else.

"Reliability"; "In contrast, a robust web of interconnected networks,
with substantial local peering at IXPs, gives networks many ways to
reach each other."

Unless you are buying transit from the others members in the IXP, then
I don't see how joining the IXP improves reliability? Indeed, the
added complexity of adding another peering point potentially could
reduce uptime.

John