Re: State of the graph

[Andrew Gallagher]

On Tue, 2021-12-14 at 18:53 +0100, Marcel Waldvogel wrote:
> Andrew,
> 
> thanks for the visualization!

NP, it's mostly Gunnar's code. He has a much more extensive history of
the mesh going back nearly three years now, at sks-status.gwolf.org

> I'm feeling flattered that keyserver.trifence.ch is at the center of
> the graph, but this also means it could become a single point of
> failure (and currently, sks.pyro.eu.org is trying to get back up to
> speed and has probably requested ~80k keys in past few hours from
> it).

There has been a distinct increase in churn on the network since it has
become more heterogeneous, and SKS<->Hockeypuck interfaces seem to be
more susceptible to churn than others.

It should be noted that pyro does not suffer from lag, but it appears
cyan on the graph because it only generates stats once per day and the
spider always seems to catch it at a bad time (hence the "?"). Also NB
that fleetstreetops and hnet.se often look like they haven't peered
back with anyone; this is because they are load balanced and only
report their peers 1 in every N scans.

> The green network looks way too much like a star topology with
> keyserver.trifence.ch at the center.

There is a definite centre of gravity to the graph, but it's not quite
as centralised as that! :-) pgpkeys.eu and cyberbits are pretty well
connected also; even if two of those three fell over it wouldn't be
enough to split-brain the mesh.

More worrying are the near-orphaned SKS nodes at the periphery of the
network that remain connected to the core only via other SKS nodes that
are in sync failure. These have been getting a slow trickle of new keys
despite being effectively disconnected; presumably the people
submitting those keys configured their default keyserver back in the
day and are unaware of the apocalypse.

>  It would be great if each of the green (and cyan) nodes could
> connect to one or two of the other nodes with low degrees. Feel free
> to use keybath.trifence.ch, but please also try to get connectivity
> which does not rely on me.

To which I'll add the usual caveat: don't silently add someone as a
peer without asking them to peer back; otherwise it won't work. :-)

A

signature.asc
Description: This is a digitally signed message part