Re: PG version recommendation - Mailing list pgsql-general

I would find out if the IT team who will maintain the system are running
a specific Linux distribution, such as RHEL and just go with the PG
version that is on that distribution.

Large corp rarely have sufficient IT resources. Unless you specifically
need a particular PG version (which does not seem to be the case based
on your info), you are better off sticking with the version provided by
whatever distro they use. This will ensure reasonable updates and
patches. In corp environments, where IT resources are thin on the
ground, any custom install often results in poor patching and update
cycles because it falls outside 'standard procedures'.

With respect to hardware specifications, it really depends a lot on what
the infrastructure is. Typically, you would be better off specifying the
performance and storage (size) you require and leave it to the IT team
to work out how to best satisfy those requirements e.g. support x
concurrent connections, y Tb/Gb of storage, backup requirements i.e. how
frequent, how many versions and retention requirements. Include details
of any additional PG packages you may need/want and how/where the
database will need to be accessed from.

As you indicate the host will be a VM, it should be relatively easy to
scale up/down cpus or memory as required, unless you have special
requirements (very complex queries, very large data sets, complex data
models involving GIS, XML, etc that may exceed resources available in
their VM infrastructure).

From your description, your database sounds fairly standard with no
unusual requirements. The number of concurrent users is low and it
sounds like it may be a new application where you probably don't yet
know where performance or resource bottlenecks will be. A standard Linux
server with 16+Gb memory and a couple of Gb for storage running PG 9.6
or higher is likely to be a reasonable starting point.

It would also be a good idea to speak to the IT team and see if they
have any procedures/policies for requesting resources. Ask them what
info they need to know and then gather that. It is unlikely to help if
yuou specify hardware requirements they cannot easily support,
especially if those requirements are really just arbitrary and based on
external recommendations from people who don't know what the
infrastructure is. Nothing frustrates IT teams more than being require
to provide over specified systems which consume valuable resources that
are never used or demand custom configurations which are unnecessary and
just add to their maintenance overheads. 

Tim

David Gauthier <davegauthierpg@gmail.com> writes:

> Hi:
>
> I'm going to be requesting a PG instance supported by an IT team in a large
> corp.  They will be creating the server as a VM.  We will be loading the DB
> using scripts (perl/dbi) on linux, possibly using bulk loading techniques
> if that's required.  Queries will come from both linux and the web, but
> typically the number of concurrent users will be on the order of 10 reads,
> maybe a couple writers.  < 1T total disk, no partitioning.  I will be
> requesting PITR.
>
> I need to pick a PG version in my request.  I want something that will be
> stable and reliable while, of course, being able to perform well.  What
> would be a good choice for PG version ?
>
> Also, since the server will be a VM, are there any special
> recommendations/suggestions might I forward in the request (install
> options, tuning options, other) ?
>
> Thanks !


--
Tim Cross