[
https://issues.apache.org/jira/browse/PROTON-1911?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Robbie Gemmell updated PROTON-1911:
-----------------------------------
Description:
Testing has highlighted that the encoding of Strings within messages isn't very
efficient, due to the manner in which the output is written to the destination
buffer. By expanding WritableBuffer to allow writing a String directly, and
improving the String encoding implementation therein we can improve performance
of String encoding. Leveraging PROTON-1913 we can also provide the previous
impl as a default fallback to avoid breaking compatibility with existing
WritableBuffer implementations that lack one.
================
Original description:
While digging into performance issues in the Eclipse Hono project I noticed a
high consumption of CPU time when encoding AMQP messages using proton-j.
I made a small reproducer and threw the same profiler at it, here are the
results:
As you can see in the attach screenshot (the first is the initial run with the
current code) most of the time is consumed in EncoderImpl#writeRaw(String).
This due to the fact that is call "put" for every byte it want to encode.
The following screenshots are from a patched version which uses a small thread
local buffer to locally encode the raw data and then flush it to the buffer in
bigger chunks.
Screenshot 3 and 4 show the improve performance, but also show that the memory
consumption stays low.
was:
While digging into performance issues in the Eclipse Hono project I noticed a
high consumption of CPU time when encoding AMQP messages using proton-j.
I made a small reproducer and threw the same profiler at it, here are the
results:
As you can see in the attach screenshot (the first is the initial run with the
current code) most of the time is consumed in EncoderImpl#writeRaw(String).
This due to the fact that is call "put" for every byte it want to encode.
The following screenshots are from a patched version which uses a small thread
local buffer to locally encode the raw data and then flush it to the buffer in
bigger chunks.
Screenshot 3 and 4 show the improve performance, but also show that the memory
consumption stays low.
Summary: improve performance of String encoding (was: Performance
issue in EncoderImpl#writeRaw(String))
> improve performance of String encoding
> --------------------------------------
>
> Key: PROTON-1911
> URL: https://issues.apache.org/jira/browse/PROTON-1911
> Project: Qpid Proton
> Issue Type: Improvement
> Components: proton-j
> Affects Versions: proton-j-0.25.0, proton-j-0.28.0
> Reporter: Jens Reimann
> Priority: Major
> Labels: pull-request-available
> Fix For: proton-j-0.29.0
>
> Attachments: qpid_encode_1.png, qpid_encode_2.png, qpid_encode_3.png,
> qpid_encode_4.png, strings_encode_after.json, strings_encode_before.json
>
>
> Testing has highlighted that the encoding of Strings within messages isn't
> very efficient, due to the manner in which the output is written to the
> destination buffer. By expanding WritableBuffer to allow writing a String
> directly, and improving the String encoding implementation therein we can
> improve performance of String encoding. Leveraging PROTON-1913 we can also
> provide the previous impl as a default fallback to avoid breaking
> compatibility with existing WritableBuffer implementations that lack one.
> ================
> Original description:
> While digging into performance issues in the Eclipse Hono project I noticed a
> high consumption of CPU time when encoding AMQP messages using proton-j.
> I made a small reproducer and threw the same profiler at it, here are the
> results:
> As you can see in the attach screenshot (the first is the initial run with
> the current code) most of the time is consumed in
> EncoderImpl#writeRaw(String). This due to the fact that is call "put" for
> every byte it want to encode.
> The following screenshots are from a patched version which uses a small
> thread local buffer to locally encode the raw data and then flush it to the
> buffer in bigger chunks.
> Screenshot 3 and 4 show the improve performance, but also show that the
> memory consumption stays low.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@qpid.apache.org
For additional commands, e-mail: dev-h...@qpid.apache.org
