Re: free collaborative tools for low BW and losy connections

[Nick Hilliard <nick () foobar org>]

Joe Greco wrote on 29/03/2020 15:56:
> On Sun, Mar 29, 2020 at 03:01:04PM +0100, Nick Hilliard wrote:
> > because it uses flooding and can't guarantee reliable message
> > distribution, particularly at higher traffic levels.
>
> That's so hideously wrong.  It's like claiming web forums don't
> work because IP packet delivery isn't reliable.

Really, it's nothing like that.

> Usenet message delivery at higher levels works just fine, except that
> on the public backbone, it is generally implemented as "best effort"
> rather than a concerted effort to deliver reliably.

If you can explain the bit of the protocol that guarantees that all 
nodes have received all postings, then let's discuss it.

> The concept of flooding isn't problematic by itself.

Flood often works fine until you attempt to scale it.  Then it breaks, 
just like Bjørn admitted. Flooding is inherently problematic at scale.

>  If you wanted to
> implement a collaborative system, you could easily run a private
> hierarchy and run a separate feed for it, which you could then monitor
> for backlogs or issues.  You do not need to dump your local traffic on
> the public Usenet.  This can happily coexist alongside public traffic
> on your server.  It is easy to make it 100% reliable if that is a goal.

For sure, you can operate mostly reliable self-contained systems with 
limited distribution.  We're all in agreement about this.

> > The fact that it ended up having to implement TAKETHIS is only one
> > indication of what a truly awful protocol it is.
>
> No, the fact that it ended up having to implement TAKETHIS is a nod to
> the problem of RTT.

TAKETHIS was necessary to keep things running because of the dual 
problem of RTT and lack of pipelining.  Taken together, these two 
problems made it impossible to optimise incoming feeds, because of ... 
well, flooding, which meant that even if you attempted an IHAVE, by the 
time you delivered the article, some other feed might already have 
delivered it.  TAKETHIS managed to sweep these problems under the 
carpet, but it's a horrible, awful protocol hack.

> It did and has.  The large scale binaries sites are still doing a
> great job of propagating binaries with very close to 100% reliability.

which is mostly because there are so few large binary sites these days, 
i.e. limited distribution model.

> I was there.

So was I, and probably so were lots of other people on nanog-l.  We all 
played our part trying to keep the thing hanging together.

> I'm the maintainer of Diablo.  It's fair to say I had a
> large influence on this issue as it was Diablo's distributed backend
> capability that really instigated retention competition, and a number
> of optimizations that I made helped make it practical.

Diablo was great - I used it for years after INN-related head-bleeding. 
Afterwards, Typhoon improved things even more.

> The problem for smaller sites is simply the immense traffic volume.
> If you want to carry binaries, you need double digits Gbps.  If you
> filter them out, the load is actually quite trivial.

Right, so you've put your finger on the other major problem relating to 
flooding which isn't the distribution synchronisation / optimisation 
problem: all sites get all posts for all groups which they're configured 
for.  This is a profound waste of resources + it doesn't scale in any 
meaningful way.

Nick