Ühel kenal päeval, E, 2006-08-21 kell 15:00, kirjutas D'Arcy J.M. Cain:
> On Mon, 21 Aug 2006 14:46:05 -0400
> "Gregory Maxwell" <gmaxwell@gmail.com> wrote:
> > On 8/21/06, Alvaro Herrera <alvherre@commandprompt.com> wrote:
> > > But the confirmation that needs to come is that the WAL changes have
> > > been applied (fsync'ed), so the performance will be terrible. So bad,
> > > that I don't think anyone will want to use such a replication system ...
> >
> > Okay. I give up... Why is waiting for fsync on a fast local network
> > which takes 15us to send a message (infiniband is cheap..) an
> > unimaginable delay when we tolerate a local 8ms fsync delay on systems
> > without writeback cache?
>
> OK, that solves your problem. How about my problem where replication
> has to happen on servers in three countries on two continents and
> thousands of updates a second have to happen in less that 10ms?
For this scenario you are far better off with partitioning than
replication.
That is if your data is partitionable. But geographically distributed
data often is.
> This is
> the critical issue with replication - one size does not fit all.
> Syncronous replication, in particular, fits almost no one.
>
> My experience is that any replication needs to be based on your business
> rules which will vary widely.
>
--
----------------
Hannu Krosing
Database Architect
Skype Technologies OÜ
Akadeemia tee 21 F, Tallinn, 12618, Estonia
Skype me: callto:hkrosing
Get Skype for free: http://www.skype.com
