[
https://issues.apache.org/jira/browse/AVRO-315?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12800112#action_12800112
]
Doug Cutting commented on AVRO-315:
-----------------------------------
Scott: This is great stuff!
I note that doReadBytes(), once the buffer is exhausted, never calls
in.read(bytes, start, length). Shouldn't we bypass the buffer in this case?
This changes semantics a bit, since currently BinaryDecoder does not buffer.
For example, in DataFileReader seeks the underlying stream. So, in that case,
we'll need to be able to reset the Decoder's buffer after a seek. There could
also be places where an application reads alternately from the Decoder and its
underlying InputStream, although, at a glance, I can't find any such in the
code currently.
> Performance improvements to BinaryDecoder
> -----------------------------------------
>
> Key: AVRO-315
> URL: https://issues.apache.org/jira/browse/AVRO-315
> Project: Avro
> Issue Type: Improvement
> Components: java
> Reporter: Thiruvalluvan M. G.
> Assignee: Thiruvalluvan M. G.
> Attachments: AVRO-315-prototype.patch, AVRO-315-test.patch,
> AVRO-315.patch, AVRO-315.patch
>
>
> The forthcoming patch improves the performance of BinaryDecoder.readLong(),
> readFloat() and readDouble().
> The test-patch has a command-line program Perf in org.apache.avro.io in the
> (test part of the source directory) which tests the performance of readInt()
> (which calls readLong()) readFloat() and readDouble(). On my machine, the
> patch improves the performance by 10% for readInt() and about 50% for
> readFloat() and readDouble().
> The idea is to unroll the loops in readLong(), readFloat() and readDouble().
> There is a small change in doReadBytes() which checks for most common
> condition before less common ones.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.