Re: [Postgres-xc-general] Info about permormance and high availability of postgres-xc

[Mason S. <ms...@tr...>]

On Sun, Oct 19, 2014 at 9:34 PM, Michael Paquier <mic...@gm...>
wrote:

>
>
> On Sat, Oct 18, 2014 at 5:12 PM, Santiago Ibarbia <
> san...@gm...> wrote:
>
>> Hi, i'm studying informatics and we are doing an investigation of
>> postgres-xc.
>> I searched information about that everyware and it isn't enough for
>> making a presentation.
>> Can you give me a little more information or a document in where I could
>> see more technical specifications??
>>
> You may find the Holy Grail here:
> http://sourceforge.net/projects/postgres-xc/files/Presentation/
>

Just to provide a bit more background information to save some time for
TL;DR cases-

Performance:

OLTP: coordinators add about 30% overhead. For linear scalability, we would
expect 10 nodes to have about 7x performance compared to vanilla
PostgreSQL. Instead it is 6 - 6.4 in DBT-1 tests, a TPC-W like benchmark.

OLAP: it depends on your schema, but if denormalized and quite simple, you
can expect pretty much linear scalability. With a more complicated schema
or queries, more nodes may perform worse than native PostgreSQL.

HA:

There is no automatic failover. You can set up standbys of key cluster
components, and fail them over manually. An alternative is to write your
own monitoring and failover, or use Corosync/Pacemaker. If acceptable in
your environment, it may even be preferable to manually failover to prevent
false positives and having to deal with failback.

I hope that helps.


-- 
Mason Sharp

TransLattice - http://www.translattice.com
Clustered and Distributed Database Solutions