Re:PostgreSQL Columnar Store for Analytic Workloads
by mason55 on Thursday April 03, 2014 @01:36PM
(#7524526)
I'd be interested to see more benchmarks. The improvements in this post are not anywhere close to what we've seen going from PG to Infobright for our reporting queries - we get speedups from 10x - 1000x, the one speed benchmark they have here is only 2x.
[ Parent
| on HN ]
Re:PostgreSQL Columnar Store for Analytic Workloads
by ozgune on Thursday April 03, 2014 @01:43PM
(#7524591)
(Ozgun from Citus Data here.)
We're going to publish more comprehensive results in a few weeks. In our initial tests, we found the speed up to depend a lot on the underlying data and the type of queries.
For example, we found that when compression reduced the working set from being on disk to in-memory, there was a significant jump.
Also, we're looking to do optimizations on the cost estimation side -- these will notably help with queries that join multiple tables together, a common scenario for the TPC-H benchmark.
Re:PostgreSQL Columnar Store for Analytic Workloads
by mason55 on Thursday April 03, 2014 @01:46PM
(#7524608)
Awesome, can't wait to see more benchmarks. I would love to be able to switch back to 100% Postgres. IB gives us tons of speed but even ignoring the MySQL warts it just feels incomplete. And we ran into a bug in production where the order of AND was actually having an effect on query results, something that's completely unacceptable and makes me worry about my data.
Re:PostgreSQL Columnar Store for Analytic Workloads
by antonmks on Friday April 04, 2014 @02:38AM
(#7529700)
Doesn't look like it uses any kind of vectorized execution. So the only advantage one gets is reduced amount of data to process. It also doesn't mention execution on compressed data so it has to decompress the data which doesn't come cheap.
Re:PostgreSQL Columnar Store for Analytic Workloads
by erichocean on Friday April 04, 2014 @10:08AM
(#7531481)
A relatively easy way to change that would be to use Intel's ISPC compiler.
Re:PostgreSQL Columnar Store for Analytic Workloads
by alecco on Thursday April 03, 2014 @02:21PM
(#7524886)
Because they still use PostgreSQL engine, very slow.
There is a great talk by Stonebraker about this. (He created PostgreSQL about 3 decades ago and moved on to newer database engines).
Re:PostgreSQL Columnar Store for Analytic Workloads
by jeffdavis on Thursday April 03, 2014 @05:57PM
(#7527213)
He started Postgres before it became PostgreSQL (i.e. before it even supported SQL). It's quite a different system now.
Re:PostgreSQL Columnar Store for Analytic Workloads
by jeltz on Friday April 04, 2014 @10:36AM
(#7531628)
The PostgreSQL Stonebraker left was a very different one than the current. It did not even have MVCC.
Re:PostgreSQL Columnar Store for Analytic Workloads
by Alex3917 on Thursday April 03, 2014 @03:22PM
(#7525444)
Is this talk online?
Re:PostgreSQL Columnar Store for Analytic Workloads
by garenp on Thursday April 03, 2014 @07:17PM
(#7527768)
Probably referring to this:
http://blog.jooq.org/2013/08/24/mit-prof-michael-stonebraker...