[
https://issues.apache.org/jira/browse/CXF-1916?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12648361#action_12648361
]
Daniel Kulp commented on CXF-1916:
----------------------------------
The proposed implementation doesn't quite work in all cases, but I think I have
it updated to work OK and I should have it committed soon (providing all the
tests pass).
Basically, the stuff that didn't work:
1) The "read()" method didn't do any boundary checks. Thus, if something used
that method, we'd have problems. Just forwarded that into the read(byte[],
0, 1) call;
2) When calling read(byte[], int off, int len), if the "len" was always less
than the boundary length, there was no way to detect if a boundary was reached.
It would just loop returning 0. Now, if the len < boundary detection
requirements, it uses an internal buffer and will pushback excess. This is
actually a bit slower than the old implementation if the "len <= 6", but that
shouldn't be a "normal" case I don't think.
> Mtom Performance - MimeBodyPartInputStream performs slow
> --------------------------------------------------------
>
> Key: CXF-1916
> URL: https://issues.apache.org/jira/browse/CXF-1916
> Project: CXF
> Issue Type: Improvement
> Components: Core
> Affects Versions: 2.0.7
> Environment: Windows XP, Java SE 1.6.0_06
> Reporter: Mayank Mishra
> Assignee: Daniel Kulp
> Priority: Minor
> Attachments: MimeBodyPartInputStream.java, TestMBPIS.java
>
>
> MimeBodyPartInputStream doesn't implements read(byte[]), hence it delegates
> it to parent InputStream class. InputStream's read(byte[]) runs over a for
> loop, eventually delegating to PushBackInputStream, which reads a single byte
> and performs boundary matching over that Byte.
> A simple test program which takes the time taken by the
> MimeBodyPartInputStream to read from a loaded buffer shows that
> MimeBodyPartInputStream takes ~1200-1400 ms to read a 12MB buffer. On the
> other hand, an InputStream takes around ~100-150ms to do so. I can understand
> that MimeBodyPartInputStream has the logic of detecting a probable boundary
> and it performs multiple if checks to do so, also it calls read and unread
> over PushBackInputStream. But counting the reads and unreads also, the time
> of MimeBodyPartInputStream looks poor.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
