On 8/20/2004 5:09 AM, dan wrote:

> Hi,
> 
> I'm a newbie to slony1 as well, but it seems to me that it's designed
> more for a round-robin type clustering system than beowulf-type
> clustering. That is, each system replicates the entire database on its
> local storage rather than handling a small part of larger queries that
> are spread over the whole cluster, and all the hosts are read-only.

Not quite, it is designed as a cascaded master->multislave system with 
support for controlled master role transfer and failover.

> 
> The tablespaces functionality in Postgresql 8 by Fujitsu may be a step
> towards what you're looking for - it allows for different tables to
> reside on separate file-systems or hosts and be transferred amongst
> these different locations without interrupting service.

I don't think that tablespaces are designed for that either. As far as I 
understood they are a mechanism to easier distribute the data and 
thereby the IO over multiple storage subsystems.

> 
> One thing I would suggest to get clustering-type effects is that you
> have INSERTs/UPDATEs only go to the master host, but queries are
> round-robined. Writes are generally much more rare than reads, so this
> allows you to have the load-balancing of a cluster without the expense
> of Oracle ;-)

The thing I would suggest is to join the multimaster replication 
development team that I intend to bring together any time soon now :-)


Jan

-- 
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck at Yahoo.com #