Hacker News Reader

Comment by slagfart | Hacker News Reader

Actually, the benchmarks aren't against alternative columnstore DBs, which would really put proof on their claims. Would love to see a comparison with Greenplum or Redshift.

For those who haven't worked with them before, columnstore databases are clever, but also do as much processing as they can at ingestion time, so that they can have good SELECT benchmarks later.

View on HN

Vertica is a pretty large commercial column store with a lot of customers. Although I'm not sure if the Yandex folks got permission to publish benchmarks about it...

(MonetDB is a classic research columnar database, and I believe MemSQL has a columnar mode. Greenplum is missing, for some reason.)

That's what I want to know, how fast is it agaisn't Vertica.

Well, they have benchmarks! Against Vertica! Which is surprising, because I imagine Vertica's EULA forbids that.

https://clickhouse.yandex/benchmark.html#["1000000000",["Cli...]

sweet, now where can I see the projections :)

Wait, Vertica forbids publishing benchmarks? What exactly are they going to do?

Clicked one of the vote buttons while trying to hit reply, sorry if it was a down...

Most DB vendors forbid benchmarks, Oracle, IBM DB2, etc, they claim they can remove your rights to run the software if you publish them

It looks like they benchmarked against MemSQL 3.2, which is nearly 2 years old and Hive .11, which is close to 3 years. Not sure how out of date the other comparisons are.

Isn't MonetDB a columnar store? They did benchmark against MonetDB.

Edit: MemSQL too is a columnar store

I stand corrected!

Edit: However, the reference states "At this time (May 2016), there aren't any available open-source and free systems that have all the features listed above". Greenplum does actually implement all of these features, and is open source. I guess I just take issue with all the hyperbole in use.

Link to greenplum db src code: https://github.com/greenplum-db/gpdb

You managed to check all of their docs already? Because as far as I know greenplum doesn't have incremental aggregation.

"Incremental Aggregation" is just UPDATE (new value = old value + JOIN value), right? If so, yes :)

Yeah... do you have ANY experience with column stores?

Actually not all column store databases do work on the ingestion path. Cassandra, HBase etc being notable exceptions. But then again not quite the same use case as the others you mentioned.

This really is begging for a better benchmark.

I thought Cassandra was all about doing work in the ingestion path. Tables are built to specify very narrow access patterns. Data is streamed to disk in a format that's very efficient to read back later (SSTable).

Cassandra is not a column store.

Reply on news.ycombinator.com