[CASSANDRA-8464] Support direct buffer decompression for reads

• Type: Improvement
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 2.2.0 beta 1
• Component/s: Legacy/Local Write-Read Paths
• Labels: performance
• Assignee: T Jake Luciani
• Reporter: T Jake Luciani

Authors

T Jake Luciani

Reviewers

Branimir Lambov

• Created: 11/Dec/14 21:13
• Updated: 29/Jan/20 12:22
• Resolved: 16/Jan/15 14:41

Description

Currently when we read a compressed sstable we copy the data on heap then send it to be de-compressed to another on heap buffer (albeit pooled).

But now both snappy and lz4 (with CASSANDRA-7039) allow decompression of direct byte buffers. This lets us mmap the data and decompress completely off heap (and avoids moving bytes over JNI).

One issue is performing the checksum offheap but the Adler32 does support in java 8 (it's also in java 7 but marked private?!)

This change yields a > 10% boost in read performance on cstar. Locally I see upto 30% improvement.

http://cstar.datastax.com/graph?stats=5ebcdd70-816b-11e4-aed6-42010af0688f&metric=op_rate&operation=2_read&smoothing=1&show_aggregates=true&xmin=0&xmax=200.09&ymin=0&ymax=135908.3

Issue Links

• is blocked by: CASSANDRA-7039 DirectByteBuffer compatible LZ4 methods
• is related to: CASSANDRA-9096 Improve ByteBuffer compression interface
• relates to: CASSANDRA-15531 Improve docs on disk_access_mode, specifically post CASSANDRA-8464