[afnog] Survey: Collecting African IXP Colocation Data for research purposes (Slight changes in the survey)

[Nishal Goburdhan]

On 1 Mar 2016, at 4:13, Roderick wrote:
> On Feb 29, 2016, at 4:45 PM, Nishal Goburdhan 
> <nishal at controlfreak.co.za> wrote:
>
> Ok I see. This might happen. However, if it does A can infer that B is 
> peering with C (C is not peering with A but is present at the IX). And 
> similarly C can infer that A is peering with B.

A can infer all it wants;  if B is operating a well run network, it will 
not change A’s view of the network.
my point was that your premise of 3 networks, and hence there will 
always be two peers to each, is incorrect, even in start-up IXPs.

note:  i’m not commenting on best practice;  i’m simply trying to 
get you to see that there are other things that you have not counted 
for, in your model.


> Could you please tell me how often 2 ASes in the region exchange part 
> of their network prefixes via an IXP although the rest is operational?

quite often actually.  what usually happens is:
* engineer A installs a peering router and connects to the domestic IXP
* to achieve some traffic engineering, engineer B advertises a more 
specific prefix to the transit operator, but forgets to do the same to 
the domestic IXP
* some sort of asymmetric routing happens

or ..

* network gets new address block/client
* network does not advertise new network block to the IXP

or ..
…

there are a few operators here that routinely complain on various IX 
lists, about inconsistent announcements that make the IX is less 
effective that it should be.  perhaps they’ll speak up with real world 
data for you    :-)


> Sure. (PS: the *academic world* finds analyzing randomly collected 
> measurements data more accurate than looking on the traffic graph 
> published on the website ;) )

IANAA.    :-)
and, for operators, what’s in the network, trumps what’s in some 
textbook.

—n.