OpenSSL without APR experiment

[Rémy Maucherat <re...@apache.org>]

Hi,

As part of a university projects, two students from the University of
Neuchatel have worked on OpenSSL support without APR, effectively producing
a JNI wrapper for OpenSSL.

It works well according to my testing and is pretty much as expected (the
JNI interfaces are now much more limited in scope), but realistically it is
useful only if there is still a plan to remove the APR connector. If not,
the amount of native code that needs to be maintained would remain the same.

The experimental code is located here:
https://github.com/jocelynthode/tomcat-native2/tree/TCN2_UNDERCAT
[If trying it, be careful about the hacking of the JNI interfaces JAR, the
readme must be followed]

Comments ?

Rémy

Re: OpenSSL without APR experiment

[Rémy Maucherat <re...@apache.org>]

2016-06-27 18:27 GMT+02:00 jean-frederic clere <jf...@gmail.com>:

> On 06/27/2016 06:07 PM, Rémy Maucherat wrote:
> > 2016-06-27 17:52 GMT+02:00 Mark Thomas <ma...@apache.org>:
> >
> >> Is the performance any different from NIO[2] + tc-native 1.2 + OpenSSL?
> >> If I recall correctly, when I asked previously you said APR/native
> >> generally had better performance.
> >>
> >
> > The performance is the same as before, there's really nothing that was
> > found and could be magically optimized (well, maybe that thread error
> state
> > then). APR remains slightly faster last time I tested, but my
> benchmarking
> > is not very legitimate.
>
> I had a huge difference in latest tests between the native connector and
> the NIO/NIO2 with openssl basically APR looked to have a problem with
> tomcat 8.5.x I didn't had to investigate.
>
> I am planning to redo tests soon... And put my slides online :D
>
> In addition to the SSL testing, non SSL testing is useful too since APR
was always a bit faster there for me [although not in a very relevant way].

Rémy

Re: OpenSSL without APR experiment

[jean-frederic clere <jf...@gmail.com>]

On 06/27/2016 06:07 PM, R�my Maucherat wrote:
> 2016-06-27 17:52 GMT+02:00 Mark Thomas <ma...@apache.org>:
> 
>> Is the performance any different from NIO[2] + tc-native 1.2 + OpenSSL?
>> If I recall correctly, when I asked previously you said APR/native
>> generally had better performance.
>>
> 
> The performance is the same as before, there's really nothing that was
> found and could be magically optimized (well, maybe that thread error state
> then). APR remains slightly faster last time I tested, but my benchmarking
> is not very legitimate.

I had a huge difference in latest tests between the native connector and
the NIO/NIO2 with openssl basically APR looked to have a problem with
tomcat 8.5.x I didn't had to investigate.

I am planning to redo tests soon... And put my slides online :D

Cheers

Jean-Frederic

> 
>>
>> If the performance was comparable with APR/native I'd have no objection
>> to dropping APR/native.
>>
>> Yes, that makes sense, we'll continue investigating.
> 
> R�my
> 

Re: OpenSSL without APR experiment

[Rémy Maucherat <re...@apache.org>]

2016-06-27 17:52 GMT+02:00 Mark Thomas <ma...@apache.org>:

> Is the performance any different from NIO[2] + tc-native 1.2 + OpenSSL?
> If I recall correctly, when I asked previously you said APR/native
> generally had better performance.
>

The performance is the same as before, there's really nothing that was
found and could be magically optimized (well, maybe that thread error state
then). APR remains slightly faster last time I tested, but my benchmarking
is not very legitimate.

>
> If the performance was comparable with APR/native I'd have no objection
> to dropping APR/native.
>
> Yes, that makes sense, we'll continue investigating.

Rémy

Re: OpenSSL without APR experiment

[Mark Thomas <ma...@apache.org>]

On 27/06/2016 16:27, R�my Maucherat wrote:
> Hi,
> 
> As part of a university projects, two students from the University of
> Neuchatel have worked on OpenSSL support without APR, effectively producing
> a JNI wrapper for OpenSSL.
> 
> It works well according to my testing and is pretty much as expected (the
> JNI interfaces are now much more limited in scope), but realistically it is
> useful only if there is still a plan to remove the APR connector. If not,
> the amount of native code that needs to be maintained would remain the same.
> 
> The experimental code is located here:
> https://github.com/jocelynthode/tomcat-native2/tree/TCN2_UNDERCAT
> [If trying it, be careful about the hacking of the JNI interfaces JAR, the
> readme must be followed]
> 
> Comments ?

Is the performance any different from NIO[2] + tc-native 1.2 + OpenSSL?
If I recall correctly, when I asked previously you said APR/native
generally had better performance.

If the performance was comparable with APR/native I'd have no objection
to dropping APR/native.

Mark


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@tomcat.apache.org
For additional commands, e-mail: dev-help@tomcat.apache.org