Re: RFC: Very large scale postgres support

On Sun, Feb 08, 2004 at 09:20:07PM -0500, Rod Taylor wrote:
> On Sun, 2004-02-08 at 21:01, Alex J. Avriette wrote:
> > On Sun, Feb 08, 2004 at 08:01:38PM -0500, Rod Taylor wrote:
> > 
> > > Replication won't help if those are all mostly write transactions. If a
> > > small percentage, even 1% would be challenging, is INSERTS, UPDATES or
> > > DELETES, master / slave replication might get you somewhere.
> > 
> > There is no way on earth we could be doing writes at that rate. I think
> > that's a given.
> 
> Sure you can, if you can horizontally partition the data so clients A
> are on machine A, clients B are on machine B, ...

I think you were assuming inserts here. The problem actually comes from
updates here. The problem is, if I update here, how long before the
rest of my "cloud" of postgres nodes understand that record is
updated?  With an insert, the transaction and propagation are fairly
clear. With an update, the overall cost is higher, and the cost
per-node is higher.

Alex

--
alex@posixnap.net
Alex J. Avriette, Unix Systems Gladiator
"You cannot invade the mainland United States. There would be a rifle behind each blade of grass." - Admiral Isoroku
Yamamoto