Skip to content

Conversation

@alamb
Copy link
Contributor

@alamb alamb commented Jul 9, 2023

Which issue does this PR close?

Related to #4973

Rationale for this change

The new grouping code is so fast that the measurement noise is becoming significant 😆 so I need larger data

What changes are included in this PR?

add tpch10 as well as tpch to bench.sh

So you can run bench.sh run tpch10 now

Are these changes tested?

I tested them manually

Are there any user-facing changes?

@alamb alamb marked this pull request as ready for review July 9, 2023 10:22
@Dandandan
Copy link
Contributor

What about increasing the number of iterations as well? This also reduces the variance in the results.

@alamb
Copy link
Contributor Author

alamb commented Jul 9, 2023

What about increasing the number of iterations as well? This also reduces the variance in the results.

That is a good idea 🤔 It turns out that DataFusion blew up my machine's memory when trying to run TPCH on SF10 as well. Maybe I also need to limit the memory (so the queries error rather than swap the OS to death)

@alamb alamb changed the title Minor: Add TPCH scale factor 10 to bench.sh Minor: Add TPCH scale factor 10 to bench.sh, use 10 iteration Jul 10, 2023
@alamb
Copy link
Contributor Author

alamb commented Jul 10, 2023

What about increasing the number of iterations as well? This also reduces the variance in the results.

Changed to 10 in bd6812d

@Dandandan Dandandan merged commit efc3827 into apache:main Jul 11, 2023
@Dandandan
Copy link
Contributor

Thanks @alamb

@alamb alamb deleted the alamb/tpch_10 branch July 11, 2023 16:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants