3 ms·

It would be great to understand why OmniSciDB does so well on this benchmark but seems to do far less well on others. The ClickHouse team was (obviously!) very

It would be great to understand why OmniSciDB does so well on this benchmark but seems to do far less well on others. The ClickHouse team was (obviously!) very interested in Mark's result and tried out OmniSciDB on the standard analytics benchmark that CH uses to check performance. Results are here: https://presentations.clickhouse.tech/original_website/benchmark.html#[%2210000000%22,[%22ClickHouse%22,%22OmniSci%22],[%220%22,%221%22,%222%22] https://presentations.clickhouse.tech/original_website/bench...] Anyway, really intriguing results from Mark. Looking forward to learning more about the source of the differences. Disclaimer: I work at Altinity, which supports ClickHouse. Edit: Fixed bad link

hodgesrm 6y ago
(To make this link work press the "100 mln." button when you arrive. It seems there is an encoding issue on this link that causes it not work correctly when copied to some sites.)
chicagobuss 6y ago
Agreed. I'm also a bit frustrated that he never tuned vertica in his test of it - he just loaded the data into the default superprojection and queried it like that. Nearly all of vertica's power comes from its concept of projections - just like normal DBs benefit from indexes. I'd really like to see how it performs to these other systems once it's been tuned properly because in my experience it's always been the fastest DB I've ever worked with. https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/Glossary/Superprojection.htm https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/Gl... https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/AdministratorsGuide/Projections/WorkingWithProjections.htm https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/Ad...
lmeyerov 6y ago
You may find the recent RAPIDS open source community GPU stack benchmark on TPCx-BB relevant, where they beat out the rest of the tools at something like 2x cheaper hw for 44x faster. It's an interesting industry benchmark b/c requires handling diverse use cases and out-of-core optimizations (eg, 1TB going through a 16GB GPU, and combo of matrix, tables, etc), so just say BlazingSQL or cuML won't really work, but all together covered it.

hodgesrm 6y ago

(To make this link work press the "100 mln." button when you arrive. It seems there is an encoding issue on this link that causes it not work correctly when copied to some sites.)

chicagobuss 6y ago

Agreed. I'm also a bit frustrated that he never tuned vertica in his test of it - he just loaded the data into the default superprojection and queried it like that. Nearly all of vertica's power comes from its concept of projections - just like normal DBs benefit from indexes. I'd really like to see how it performs to these other systems once it's been tuned properly because in my experience it's always been the fastest DB I've ever worked with. https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/Glossary/Superprojection.htm https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/Gl... https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/AdministratorsGuide/Projections/WorkingWithProjections.htm https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/Ad...

lmeyerov 6y ago

You may find the recent RAPIDS open source community GPU stack benchmark on TPCx-BB relevant, where they beat out the rest of the tools at something like 2x cheaper hw for 44x faster. It's an interesting industry benchmark b/c requires handling diverse use cases and out-of-core optimizations (eg, 1TB going through a 16GB GPU, and combo of matrix, tables, etc), so just say BlazingSQL or cuML won't really work, but all together covered it.