Optimizations possible

1. Inplace sorting of fixed types

At the moment we do an index sort. Inplace sorting is much more efficient than index sorting when possible. There might be algorithms capable of in-place sorting variable-length records. We need to do research on this.

What is the impact of in-place sorting on variable-length records in a table? We can create an index while doing in-place sorting of the possible columns and then use the index to do variable lengths.

2. Merge sorting and computation + communication overlapping

We can do merge sorting for overlapping computation + communication. If we do in-place sorting, we can just combine the chunks.

3. Partition to cache sizes

We can partition the elements according to cache sizes to optimize further

4. Separate out null type handling

We need to handle null cases when sorting. We can swap the null cases to the beginning before in place sorting.