Re: network topology and stuff

[Pekka R. <pri...@ik...>]

: When the time is ready, I don't know. Pekka perhaps enlights us with
: technical stuff about backup routing.
:
As the protocol spec indicates the current backup router scheme really
isn't 100% proof system, and I'm not really even happy with the current
system.  It perhaps should be done some other way but I don't know how.

Also, one other thing that the backup router scheme does not take into
consideration is the netsplits inside the cell.  Meaning, when for example
server connection to the router is lost due to routing problems or very
long network lag.  Currently, it's a netsplit between that server and
router.

I however, would like to see more resilient system where we would just say
that "I don't accept the fact that the transport was lost to the
server/router and we split".  If it is lost, then just get it back as soon
as possible, but do not split.  So, if server looses connection to its
router (and vice versa) due to network problems, the server/router
wouldn't send SERVER_SIGNOFF like it does now.  Instead, it would just
do everything in its power to get the transport back.  Messages would be
cached while transport is lost, etc.  It would be like "opportunistic
netsplit avoidance".  If I split, then accept the fact that you've lost
the transport, but do not accept the fact that SILC network would split.
Instead, just repair the transport and be like nothing happened...
Naturally, if the server/router crashes then this has no meaning, but
that's different...  If the transport repairing takes too much time, then
it would split (SERVER_SIGNOFF).

I actually scheched around with this scheme while I was doing the new
protocol specs but decided to leave it for later time.  What I really
would like to see is that someone would actually start figuring out this
by themselves and not expect that I will deliver it at some point, and
this goes for the backup router stuff too...

: There have been offers for new servers from many people around the world
: already, some of them will propably want to run a router or backup
: router already.
:
It's already been demonstrated that bringing new servers don't bring new
users.  Shouldn't the network expansion be based on the true need to
expand due to user growth?  Or is the large but fairly empty network where
there is much servers and routers better?  And this wasn't a retorical
question, but real question.

: countries. Instead one cell should contain servers from multiple
: locations, countries and continents. Correct me if I've understood the
:
My plan has been than for starters it would really be nice to have one
router per continent, but naturally taking into considreation the fact
that some continents are bound to have more users than others.  Dunno...

: Also some kind of SilcNET organization should be set up to think and
: discuss about network topology, linking policies, admin behaviour
:
Yes, this would be fine.  Currently the pond is pretty small for lot of
people though...

	Pekka
________________________________________________________________________
 Pekka Riikonen                                 priikone at silcnet.org
 Secure Internet Live Conferencing (SILC)       http://silcnet.org/