I have to implement product independent database replication and would
appreciate any hints on some theories regarding replication.

What I want (have) to do:

- several databases on distributed servers
- no master database
- changes are made to any of the distributed databases
- the changes have then to be synchronized on all databases

My customer wants a detailed desription of the used replication mechanisms
and the source code that is doing this. So I can not use any commercial
product. Furthermore the mechanism should potentially work with any
available database without major changes in the replication software. (For
the moment the databases will be ORACLE and InterBase).

I currently have two ideas how this could be realized:

1) Create my own "transaction log" in a database table and use that for
replication
2) But a timestamp on each record of every table in my database

So my questions:

Does anyone know if 1) or 2) would be a good mechanism to use?
Can anyone give me any pointers to (detailed) information regarding
replication algorithms?

Thanks for any help!
Daniel

Daniel wrote:

Not if you want to guarantee integrity and consistency.

A good starting point is the chapter on distributed databases in Date's
"An Introduction to ...". That's chapter 20 in the 7th edition, and
chapter 21 in the 8th edition. Especially the parts on recovery control
where he discusses the two-phase commit protocol are relevant. It is
this protocol that you should probably be using.

-- Jan Hidders