[TAJO-36] Improve ExternalSortExec with N-merge sort and final pass omission

• Type: Improvement
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 0.8.0
• Component/s: Physical Operator
• Assignee: Hyunsik Choi
• Reporter: Hyunsik Choi
• Created: 18/Apr/13 08:13
• Updated: 04/Feb/14 21:14
• Resolved: 04/Feb/14 21:10

Description

Background:

The current ExternalSortExec just uses the binary external merge sort algorithm (http://en.wikipedia.org/wiki/External_sorting#External_merge_sort). In other words, for each pass, ExternalSortExec just merges two files into one sorted file.

Proposal:
The goal of this proposal is to improve ExternalSortExec with the following improvements:

• N-merge sort - we can merge N files though more memory at each pass. It will reduce the number of passes. Consequently, it will reduces considerable I/O overheads.
• the final pass omission - a physical operator is pipelined by the parent operator. The final pass of the merge sort must also be invoked by the parent physical operator. So, we can omit the final pass of the merge sort.

Issue Links

• is depended upon by: TAJO-584 Improve distributed merge sort