Re: Vertica targeting PostgreSQL users

On Nov 20, 2017, at 6:52 PM, Peter Geoghegan <pg@bowt.ie> wrote:

On Mon, Nov 20, 2017 at 7:43 AM, Jonathan S. Katz <jkatz@postgresql.org> wrote:

Vertica current marketing is heavily targeting Postgres users:

https://www.vertica.com/postgresql/

Going forward, we must continue to understand our users’ needs while
ensuring that we can provide them as many resources as possible to help them
manage Postgres and show them that here is great help available when it is
required.

I noticed this quote, which was fairly prominently placed:

"PostgreSQL just isn’t designed to do analytic-type queries. Those are
big aggregations, and Postgres, even though it is a great relational
database, is really tailored for single-record lookup."

Isn't the latter sentence really quite fair? It seems unwise to try to
compete with a dedicated MPP column store solution like Vertica. Of
course it's going to be much better at Postgres for a use-case that is
truly within its niche.

That said, I definitely think that systems like Vertica could easily
get users due to the extremely simplistic, over-confident thinking
that many people display around scalability. For some reason, I've met
a number of people that believe that using Postgres somehow becomes
untenable once you reach 1TB of data. Ideas like this imbed themselves
by being simple, and getting repeated without being challenged. People
think they need a column store, or something like Cassandra, when in
fact they need to do some performance triage using pg_stat_statements,
rethink backups, and maybe upgrade hardware.

The lesson for us, as people that want to do better advocacy, may be
that we need to counter these preposterous rules of thumb with simple
counter examples.

For example: I can restore the entire stack overflow
databases on my laptop; it contains all stack overflow posts, ever,
and comes in at approximately 100GB, including basic indexes. The
largest table can have an index created on it in a couple of minutes
on my machine that weighs less than 2KG, as we see here:

https://blog.anayrat.info/en/2017/11/19/postgresql-10--icu--abbreviated-keys/

The actual stack overflow production database runs on a single SQL
Server node, and does come in at 2 - 4 TB IIRC, because they have
event data too, but the fact remains that you essentially get all of
stack overflow in a ~100GB Postgres database. Many users have no idea
how far a traditional monolithic relational database can scale without
much difficulty, because they measure the wrong thing -- the thing
that is easiest to measure.