On Tue, 2010-02-09 at 16:53 -0500, Marc Fournier wrote: > > > pgcluster-I design is not scalable. You will need *at least* 3 > cluster > > machines and 1+1 load balancer + 1+1 replicator machines for a > system > > that is won't stop (at least for a while) > > Please elaborate on this point ... you say you will need *7* machines, > total? Why?
You need at least 3 db servers. Current design is: If one of them fails and comes back, it will need to take another host down, and re-sync from it -- and 3rd machine will have all the load. For the load balancer and replicator: You need to have a backup for both. You *can* install load balancer to one of the db machines, but it is will be a SPOF. I don't find installing load balancers to db servers clever, because of overhead and such. Same is valid for replicator instance. > > Until PostgreSQL developers implement some way to share > "shared_buffer" > > data between more than 1 PostgreSQL machines, none of the solutions > will > > be scalable. PGCluster does have lots of overhead, including write > > loads. The only gain is from SELECT queries. However, given the > > complexity of the installation and infrastructure, I would not spend > > more time on PGCluster on prod installations. > > Odd, I found the installation process for 1.9 to be very > straightforward and easy ... once I built/installed the binaries, it > took me <15 minutes to get it configured and running where I could > load databases and actaully see them on both sides ... was very > impressed with that ... Did you test it under heavy load? I did. -- Devrim GÜNDÜZ Professional PostgreSQL Services, Training, 24x7 Support Authors: PostgreSQL RPMs, PITRTools, PostgreSQL Replicator http://www.commandprompt.com/ XMPP: [email protected] Twitter: http://twitter.com/devrimgunduz
signature.asc
Description: This is a digitally signed message part
_______________________________________________ Pgcluster-general mailing list [email protected] http://pgfoundry.org/mailman/listinfo/pgcluster-general
