cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jonathan Ellis (JIRA)" <>
Subject [jira] Commented: (CASSANDRA-408) Pool BufferedRandomAccessFile objects used by sstable reads
Date Wed, 30 Dec 2009 04:03:29 GMT


Jonathan Ellis commented on CASSANDRA-408:

Patches attached to perform mmap-backed reads.  Crappy testing on my laptop shows about a
15% speed increase w/ reads.

Old read path is still around, primarily for use on 32bit systems.  Old path is also used
on rows that cross the 2GB boundaries that the JVM lets us map at a single time (boo!) and
for compactions.

> Pool BufferedRandomAccessFile objects used by sstable reads
> -----------------------------------------------------------
>                 Key: CASSANDRA-408
>                 URL:
>             Project: Cassandra
>          Issue Type: New Feature
>          Components: Core
>            Reporter: Jonathan Ellis
>            Assignee: Chris Goffinet
>             Fix For: 0.9
>         Attachments: 0001-add-FileDataInput-implemented-by-BufferedRandomAccessF.txt,
0002-Implement-FileDataInput-with-MappedFileDataInput-backe.txt, 0003-productize-mmap-approach-handle-files-2GB-by-chunking-.txt,
408.patch, commons-pool-1.5.2.jar
> not only does BRAF per op do a whole lot of extra fopens, but the buffering actually
makes it _more_ expensive to set up since on the jvm all primitive arrays are initialized
to zero.
> this adds a simple read test to; I'm seeing about a 10% increase in throughput
which is worth 200loc imo.

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

View raw message