Github user liancheng commented on the pull request:
https://github.com/apache/spark/pull/1587#issuecomment-50126452
@marmbrus has already filed
[SPARK-2650](https://issues.apache.org/jira/browse/SPARK-2650) to track this
issue and assigned to me. The `104` here is definitely a stupid typo (my bad,
sorry...). But the more important missing piece here is that we haven't
implemented the logic to calculate a proper initial buffer size like what Shark
does. Shark initializes the buffer size by estimating the size of the target
column based on DFS block size and `ColumnType.defaultSize` (see
[here](https://github.com/amplab/shark/blob/f8c16f2b6c5990279376e9fd7178301f0341c6b2/src/main/scala/shark/memstore2/ColumnarSerDe.scala#L57-L65)
and
[here](https://github.com/amplab/shark/blob/f8c16f2b6c5990279376e9fd7178301f0341c6b2/src/main/scala/shark/memstore2/ColumnarSerDe.scala#L57-L65)).
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
