replication

Christoph Dworzak (cdindex nospam at amazing.ch)
Sat, 13 Mar 1999 06:43:00 GMT

Hi again

I was thinking about the data-replication:

What I'd like to see is a group of servers which
act as equal peers. No central host or some such
thing.

Netnews works this way, but I don't like the idea
of installing a newsserver just for this :)

There's no predefined structure (neither tree nor
chain). A server-admin just agrees with another
one to exchange data. You could run it isolated
in your intranet or form a competitive group when
you're unhappy with the current one.
You could let everyone read data from your server
but accept changes only from trusted partners
or specify whatever policy you want.

Of course there are some points to define to make
this work smootly. One of course is the integrity
of the database. Submissions could happen on any
server, so replication is mostly bidirectional.
This means there could be conflicts (either 2 CD
were newly entered on different servers or changes
must be merged).
I think this is solvable. last wins, voting or
merging of the data.

Another point to solve is to decide what to
replicate. My first Idea is to datestamp all
records. Server A would ask server B for all
changes which happend since they last met.
Then they would turn around and B would send
changes to A.

One concern I heard was latency. Maybe a central
host replicating everything downstream could be
faster, but we're speaking about servers on the
internet here. They're online all the time and
there won't be x changes per second (well, never
say never, but I really don't expect it :)
If your server should be really laged, you're
free to conect to a faster one...

Bye
dworz