Begin Daniel wrote:
> On Sat, Apr 2, 2016 at 1:42 PM, JORGE MALDONADO <jorgemal1960@gmail.com> wrote:
>> I am developing an information system that will be used by several
>> clientes, each client has its own database, and each database has
>> exactly the same structure for each client. The only difference is the name of the database.
>> I have been thinking about whether or not using one tablespace to
>> create all of the databases or to use one tablespace for each
>> database, but I have not found a good reason for using one method or the other.
>
> Generally I have found it best to use a single tablespace except when there is a clear reason to do
> otherwise. Usually that reason, if it exists, is to allow storing less frequently accessed data on a
> slow, cheaper medium.
> --------
> Another reason to use multiple tablespaces is if IO operations are slowing down significantly because
> the queries have to access several large tables at the same time (i.e. read/write tens of GB by
> table). In this case, it might be a good idea to distribute IO operations by spreading on several
> disks the tables that are often used together.
>
> Otherwise (smaller tables or expensive queries are only sporadic), keep everything together as Kevin
> suggested.
I have been told that the device I/O queue can become a bottleneck on
Linux if there is a lot of I/O going to a single device, and that it is
better to spread this across moltiple devices (via tablespaces or
striping).
I have not been able to personally observe this.
Yours,
Laurenz Albe