[CASSANDRA-8923] Improve MergeIterator performance for binary prefix comparable data

• Type: Improvement
• Status: Open
• Priority: Low
• Resolution: Unresolved
• Component/s: Legacy/Local Write-Read Paths
• Labels: compaction performance
• Assignee: Unassigned
• Reporter: Benedict Elliott Smith
• Created: 05/Mar/15 16:48
• Updated: 16/Apr/19 09:31

Description

Along the same lines as CASSANDRA-8915, only designed to help merges of small partitions, or those with clustering prefixes that are binary prefix (or byte order) comparable. The approach in CASSANDRA-8915 will work well for normal intra-partition merges, but partition keys themselves are typically distributed by hash, meaning there will very rarely be runs of keys owned by a single source. A binary trie based merge would perform significantly better here, and could also be rolled out alongside CASSANDRA-8731 for binary prefix comparable fields.

Issue Links

• is related to: CASSANDRA-8731 Optimise merges involving multiple clustering columns
• is related to: CASSANDRA-11697 Improve Compaction Throughput
• is related to: CASSANDRA-8906 Experiment with optimizing partition merging when we can prove that some sources don't overlap
• is related to: CASSANDRA-8918 Optimise compaction performance for unique partition keys
• is related to: CASSANDRA-8915 Improve MergeIterator performance