Impala now faster than 'commercial analytic DBMS' ?

[peeterskris]

http://blog.cloudera.com/blog/2014/01/i ... ass-speed/

Could the 'commercial analytic DBMS' possibly be Vertica?

[JimKnicely]

I don't think so.

http://www.quora.com/Cloudera-Impala/Ho ... to-Vertica

[scutter]

I’m not sure that anything can be concluded from the Quora references. But from a quick scan of the Impala vs. DBMS-Y article, it refers to data being in a column store but says nothing about it being optimized. If it were vertica - is a comparison with unoptimized projections meaningful? I don’t think so.

—Sharon

[JimKnicely]

I was just implying that if Cloudera was using Vertica there'd be no need to make the comparison in the first place

[peeterskris]

With Parquet, Impala now has column store as well. I met the guy behind Impala, Marcel Kornacker, a while ago at a conference. They wanted to be as fast as Vertica within a year.

Now I know, wanting something, and doing your own tests, doesn't make it so. For one, HDFS is still a blocker, as far as I know. With Vertica you have control over how you segment your data. You typically replicate your small tables for efficient, local joins. Last time I checked, joins were always distributed in Impala.

Still, it's a player to keep an eye on.