[
https://issues.apache.org/jira/browse/CASSANDRA-8630?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14704551#comment-14704551
]
Stefania commented on CASSANDRA-8630:
-------------------------------------
I've rebased to 3.0 and renamed the branch to
[8630-3.0|https://github.com/stef1927/cassandra/tree/8630-3.0].
I had conflicts with CASSANDRA-6230, since it introduced
{{ChecksummedDataInput}}, a new specialization of {{AbstractDataInput}}. I
integrated it somehow but I feel there is some duplication between this class
and {{ChecksummedRandomAccessReader}}. However the two aren't identical either,
so for now they are separate.
I made good progress with moving the segments to the builders and replacing the
map with an array but this is still not complete. If you want to take a quick
look the class doing the bulk of the work is called
[MmappedRegions|https://github.com/stef1927/cassandra/commit/d1418ab889f60812cc866f12bf94b2360b3bb2d3#diff-88342f36d0687d3a0559fede5d158d83R33].
Your feedback is welcome but it is far from complete, specifically I still
need to make it into a ref counted object, since the builders don't necessarily
survive the files, and address thread safety issues. Also, I merely extend
segments at the moment, I make no effort to compact them. I did not go for the
two arrays approach since the code is more readable with a single array of a
well defined class but if you really think this makes a big difference I can
change it.
The second batch of code review comments is also still WIP. I will post more
details once it is complete.
> Faster sequential IO (on compaction, streaming, etc)
> ----------------------------------------------------
>
> Key: CASSANDRA-8630
> URL: https://issues.apache.org/jira/browse/CASSANDRA-8630
> Project: Cassandra
> Issue Type: Improvement
> Components: Core, Tools
> Reporter: Oleg Anastasyev
> Assignee: Stefania
> Labels: compaction, performance
> Fix For: 3.x
>
> Attachments: 8630-FasterSequencialReadsAndWrites.txt, cpu_load.png,
> flight_recorder_001_files.tar.gz, flight_recorder_002_files.tar.gz,
> mmaped_uncomp_hotspot.png
>
>
> When node is doing a lot of sequencial IO (streaming, compacting, etc) a lot
> of CPU is lost in calls to RAF's int read() and DataOutputStream's write(int).
> This is because default implementations of readShort,readLong, etc as well as
> their matching write* are implemented with numerous calls of byte by byte
> read and write.
> This makes a lot of syscalls as well.
> A quick microbench shows than just reimplementation of these methods in
> either way gives 8x speed increase.
> A patch attached implements RandomAccessReader.read<Type> and
> SequencialWriter.write<Type> methods in more efficient way.
> I also eliminated some extra byte copies in CompositeType.split and
> ColumnNameHelper.maxComponents, which were on my profiler's hotspot method
> list during tests.
> A stress tests on my laptop show that this patch makes compaction 25-30%
> faster on uncompressed sstables and 15% faster for compressed ones.
> A deployment to production shows much less CPU load for compaction.
> (I attached a cpu load graph from one of our production, orange is niced CPU
> load - i.e. compaction; yellow is user - i.e. not compaction related tasks)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
