Olivier Mascia wrote:

> Would shadowing (and not journalling) to a *different* server make
> sense ? A kind of master-slave remote on-line replication but at
> page/system level, not at row/application level ?
>
> Going even further, couldn't we even think (or is dream the right word ?)
> of a kind of cluster-like configuration (basic "classic server" with distributed
> lock manager), working on two replica of the DB (different physical
> locations) instead of a common one ?

As a side note, I'd also recommend us looking at the PostgreSQL's WAL.
They seem to did a pretty successful and painless implementation of it
(done by a single person*, works pretty well right away, didn't even
increase the major version number for the version that has WAL for the
first time).

* the same person who implemented MVCC in PostgreSQL.

--
dave