Just had some idea and wanting to share while I remember it. Comments welcome!
Right now storage of erasure shares is distributed by means of /24 IP blocks. We know this can be worked around by node operators to some degree by, let say, using multiple ISPs or setting up VPNs. However, this method has a crucial feature of being extremely simple and fast when selecting nodes.
I’m now imagining a network of, let say, 40 «measurement nodes» controlled by a satellite owner, and located in various parts of the world, preferably so that each node is in a different data center, different city, different geographic area. Each of them tests all nodes periodically and measures response latency (maybe as a side effect of audits, so these wouldn’t be actually new queries?). For each storage node we find the measurement node with the lowest latency and classify it as belonging to the geographic area of the measurement node.
Then, at storage node selection, nodes are selected so that no geographic area is selected more than X times, for some tunable X. The area would be identified by a single small number, one byte, so selection could again be fast.
It would be much more difficult to cheat, as it’s hard to have a single location have low latency to more than a small number of data centers. VPNs only increase latency, and multiple ISPs will likely connect to similarly-located backbone nodes anyway.
As for feasibility of running so many measurement nodes, they could be some smallest VPSes at, let say, Azure, which is already present in 54 regions. Alternatively I suspect that it should be possible to run a smaller number of measurement nodes and consider a geographic area as area defined by two lowest latencies. E.g. an area with the two lowest latencies to Paris and Frankfurt is likely different than an area with low latencies to Paris and Madrid.
While I don’t think the current approach is so bad that it must be changed, the above idea would probably allow for more fair distribution of data and take into consideration the fact that tech hubs are likely to amass large number of IP blocks.