[ixp] others at ixp - failure of ISPs?

[Mark Tinka]

On Tuesday 02 June 2009 11:14:48 pm Michuki Mwangi wrote:

> This is highly reminiscent of the Kenyan experience with
> the incumbent telco to the extent that it put in an E1
> link to one content provider (Revenue Authority Services)
> to whom they required good connectivity else they were
> loosing clients to ISPs.

Clearly, a case like this would threaten the incumbent's 
competitive advantage. The Revenue services is non-
competitor per industry, however, a lot of its customers 
would need to have decent access to them for successful 
online transactions.

It's actually in the incumbent's best interests to have good 
connectivity to such key service providers as the Revenue 
folks. In fact, either peering with them or trying to get 
them to multi-home would be something they can even consider 
as a loss leader, in the grand scheme of things.

Try turning the tables though, and the other end of the 
connection is a competing ISP. Considering that incumbents 
generally have most of the eyeballs, competing networks' 
customers complain to their provider, competing networks 
complain to the incumbent. The push for improvement is in 
the opposite direction - and that's not saying much.

> entirely dependent how clued in your Government/regulator
> is. Having the right folks lobbying within their circles
> can help smoothen the issues out. By and large maybe
> doing some policy workshops (read layer 9) around the
> issues maybe a starting point.

The policy workshops are great, but what we've seen is that 
the commerce tends to win out in the end with these 
particular issues.

Incumbents will be willing to upgrade their connectivity to 
the exchange point, as long as the government/regulator can 
fund that bit of the project as well.

My concern is how sustainable something like this, when 
priorities begin to shift. Self-sustaining exchange points 
tend to work best in the long run. If anything were to 
affect the continued flow of funds from the 
government/regulators, it would be the perfect opportunity 
for the incumbent to pull out, and force all other members 
to either pay more to their upstreams to get to them, or 
purchase expensive local peering.

> To fix the issue with the incumbent was achieved through
> formally depeering them. Then local community then
> started fussing over their inability to route traffic
> locally (on local mailing lists) and this appears to have
> done the trick in getting them back on track with right
> capacity and announcing all their prefixes.

I suppose bowing to pressure does have its success stories, 
good for .ke.

My worry about what happens when traffic profiles increase, 
asymmetrically between peers, and networks begin to feel 
they are somewhat subsidizing other networks. But this all 
depends - it's not necessarily cast-in-stone as different 
regions face different issues.

> The key is making the IXP attractive enough (with
> content) for folks to want to peer. Having key
> infrastructure in IXPs is one way to go about this i.e
> root-servers, ntp, and ccTLD available at an IXP is a
> start. I suppose talking to the likes of Google for their
> G2C nodes would be something to create interest around
> the IXP.

I think the amount of useful content you can have at an 
exchange point to make it attractive is limited, save for 
what networks are already serving up with their NLRI.

You mention the usual suspects, i.e, ccTLD, root name 
servers, e.t.c., but some exchange points have, in the past, 
suggested other content such as favorite web sites, cache 
servers, e.t.c. In practice, we've seen that this hasn't 
worked very well, if at all.

In many cases, exchange points are not geared up to handle 
massive amounts of data warehousing typical of purpose-built 
co-lo facilities.

I think increasing the number of value peers at the exchange 
point, as well as situating it around a good location where 
local, regional and international networks may interconnect 
would very likely make it more successful, all other details 
involved in running exchange points being the case, of 
course.

> I think its always useful to maintain both models with
> voluntary multilateral peering and bilateral peering.
> IMHO it means that the IXP will attract both big and
> small as it evolves as opposed to loosing most members to
> smaller IXPs that are more designed/suited for their
> peering needs.

Completely agree - "voluntary" being the word.

Heck, I'm in support of everyone connecting to everyone, in 
an ideal world. However, the economics of connectivity are 
far more complex, or perhaps, simple :-).

Cheers,

Mark.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: This is a digitally signed message part.
URL: <http://www.afnog.org/pipermail/ixp/attachments/20090603/47aa0627/attachment.bin>