Yannick Lecaillez wrote:
> Hi ppl,
>
> What need to do (understand, to devel) to allow several
> postgres instance running from several server to access to the
> same data (no replication at all) hosted on a SAN ?
>
> I'm probably wrong but i think this type of dev should
> be easier to realize than replication ? Because all node are
> always consistent since they use exactly the same data ?
The problem is the PG uses shared memory to coordinate the various
backend processes. Sharing the files is not the difficult bit, sharing
info about who is doing what with which tuple/has which locks is.
> In fact i'm asking why opensource db go only to the
> replication solution than "true" clustering solution using SAN
> like Oracle RAC ? I don't see the big advantage of the replication
> method (of course, for application which need more than few nodes
> hosting "small" db). The price could be an answer but what about a db of
> several hundred giga byte ? When a node is added it must contain this
> capacity nearly for "nothing" (i thinks its more and more a waste when
> number of node grow ...).
The advantage of replication is that you can just use standard server
boxes. There is a company offering specialised high-bandwidth hardware
that is supposed to offer PG clustering, but I'm afraid I know no more
about it.
-- Richard Huxton Archonet Ltd
