[atlas] More even earth distribution of measurement probes

Dear Marat,

On 10/03/17 09:47, Marat Khalili wrote:
> I'd like to check worldwide connectivity to my site. I create a new
> measurement via web-interface and add, say, 300 probes, expecting them
> to cover most world countries.

There was a hackathon project that made this possible in some way, 
called "Spatial Bucketing of RIPE Atlas Probes on Map Projection", by 
Julian Hammer:

"[he] wanted to solve the following problem: when scheduling 
measurements from lots of probes, requesting a "random worldwide 
selection" currently results in a choice that is very biased towards 
Western/Northern Europe and the USA (see image without Sbucket on the left).

Julian's solution makes a more equal selection, by using the grid to 
divide the globe and choose probes based on that division (see image on 
the right after applying Sbucket)."

Images: https://labs.ripe.net/Members/becha/ripe-atlas-hackathon-results

Code: https://github.com/cod3monk/RIPE-Atlas-sbucket

I hope you find this useful!

Cheers,
Vesna

> However, looking at the map
> <https://atlas.ripe.net/measurements/7861872/?filter=#map> I notice that
> most are selected from Europe: there're two dozens in Netherlands (many
> with same ASNs), but only two in Australia, one in Japan and nothing in
> Brazil. Obviously, most ATLAS probes are from Europe, and selection
> algorithm does not take probe density into account.
>
> Wouldn't it be nice to have the following features in web-interface and API:
> * even distribution of probes by country or continent based on
> population, area, and other user-defined weight numbers;
> * avoiding probes with same ASNs
> ?
>
> I don't know how current algorithm works, but easiest modification would
> be two-step process: (1) randomly select country based on specified
> weights; (2) randomly select probe within country, avoiding earlier
> selected probes or ASNs and failing if none are left available; repeat
> from step 1 until necessary number of probes is selected.
>
> P.S. I tried to manually request 1 probe from each country, but stumbled
> on unrecognised country codes, countries with no probes etc. before even
> trying population-weighted distribution. Without direct access to probes
> database nice solution can be difficult to achieve via current API,
> although crude one looks possible.
>
> P.P.S. I understand this option will create more load on probes in
> less-covered countries, but since it's possible to limit selection to
> specific countries anyway this is hardly a strong argument against
> adding this feature.
>
> --
>
> With Best Regards,
> Marat Khalili