Vernon Schryver
vjs@calcite.rhyolite.com
Wed, 16 Jan 2002 13:15:15 -0700 (MST)
> From: John Payne <john@sackheads.org>
> Is there any documentation on the server-ID path?
not yet
> And is it
> good/bad/indifferent if there are loops seen in the path?
>
> Just out of curiousity I had a look at the paths I'm seeing,
> and I see double digit counts of paths like 101<-1015<-101<-1016
A loop implies one of several things:
- a duplicate server ID, which is a Bad Thing(tm)
- server ID translating causing confusion, probably also a Bad Thing
- a server that lost its database and was restored by its neighbors
- broken or experimental software and/or "bad hands."
Last week I tried something with the authenticated client bit that
didn't work out.
> Should we also be looking to exchange floods with as many people
> as willing, or people we see lots of times in the path, or is it
> better to have a couple of big exchangers and lots of leaf servers?
I think Usenet provides the right models (plural).
In that universe,
- there are both big exchangers and lots of leaves.
- there are both full feeds and leaf-only upstream or partial feeds
- every server needs at least 2 and possibly 3 full feeds links.
- no server needs more than 3 or 4 full feeds.
The normal DCC bandwith is likely to remain small, but when servers
need to be reflooded, wires get hot. The current flood is about 4
MByte/day and the 7-day database is about 30 MByte (with perhaps a
doubling within the next month). Many 4 MByte/day streams are no
problem, even for a 56 Kbit/sec modem. However, if you have a few
dozen flooding peers and you do something that causes refreshing of
the full 30-60 MByte (current) database, things get kind of slow.
The equivalent to a netnews "L1" upstream feed is something like this in
/var/dcc/flod:
dcc.foo.com 1234 5678 self->self,all->reject
An L2 feed where server-IDs 4444 and 555 are downstream might be
dcc.foo.com 1234 5678 self->self,4444->4444,5555->5555,all->reject
Vernon Schryver vjs@rhyolite.com