[BUG] cuIO benchmarks generate larger files than expected
Open
bug
cuIO
libcudf
tests
- Dominant language
- C++
- Stars
- 9.8k
- Forks
- 1.1k
- Avg merge
- 3d 6m
- Merged PRs (30d)
- 278
Description
cuIO ORC and Parquet benchmarks generate larger files than previously (e.g. GDS blog data).
Some observations:
- [x] Only the cases where both cardinality and run length are set lead to small files.
- [ ] Cases where the data does not encode/compress well now generate files larger by ~20% (based on integral columns).
Contributor guide
Assessment
This issue has not been assessed yet.