Github user ChengXiangLi commented on the pull request:
https://github.com/apache/flink/pull/1255#issuecomment-165652344
Hi, @fhueske , For the partition part, i think it's normal that
`RangePartition` is slower than `HashParition`, as you've mentioned,
`RangePartition` introduce more overhead. The most difference between
`HashParition` and `RangePartition` is that, `HashParition` is key-wise
partition(elements with same key would shuffled to same target), and
`RangePartition` is key-wise and partition-wise partition(the partition is in
order as well), so for global order, we can sort in parallel after
`RangePartition`, that's what we can benefit from `RangePartition`.
On the other side, it's still make sense to improve `RangePartition`
performance, although i don't think increasing the sample size would help here.
Based on my previous calculation and test, `parallelism * 20` is enough to
generate well-proportioned partitions. Do you find there is data skew in any
partition after `RangePartition`?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
