Connector refactoring progress update

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

The latest series of commits is the prep work for merging the
Http11InputBuffer implementations. I'll be doing that shortly.

The current status is that the tests pass but performance is currently
worse for all connectors but especially for NIO2.

Once I complete the InputBuffer merging, I plan on spending some time
looking at tuning. I suspect that the buffering is not as efficient as
it could be.

Mark

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

Re: Connector refactoring progress update

[Rémy Maucherat <re...@apache.org>]

2015-01-15 13:21 GMT+01:00 Mark Thomas <ma...@apache.org>:

> Thanks for the pointer. That explains what I was seeing. I had
> maxKeepAliveRequests at the default so the delay was impacting the
> results. Ignore the delay and the 8.0.x vs. 9.0.x results are comparable.
>
> I'll take a look and see if I can figure out what is going on. Something
> related to a pending read I suspect.
>
> The ab sometimes completes, but it always has a handful of failed requests
(less than 5 over 1M) so there's some corruption going on. It would
probably be more obvious on other payloads. Maybe this could be something
about the awaitBytes you weren't sure about. Good thing the 8.x code can
still be used as a reference.

Rémy

Re: Connector refactoring progress update

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

On 15/01/2015 10:25, Rémy Maucherat wrote:
> 2015-01-15 11:11 GMT+01:00 Mark Thomas <ma...@apache.org>:
> 
>> The latest series of commits is the prep work for merging the
>> Http11InputBuffer implementations. I'll be doing that shortly.
>>
>> The current status is that the tests pass but performance is currently
>> worse for all connectors but especially for NIO2.
>>
> 
> I just checked and my ab doesn't complete for whatever reason on NIO2 (the
> very last request hangs), but with the right timing to kiil it, I don't
> think the throughput is very much changed.

Thanks for the pointer. That explains what I was seeing. I had
maxKeepAliveRequests at the default so the delay was impacting the
results. Ignore the delay and the 8.0.x vs. 9.0.x results are comparable.

I'll take a look and see if I can figure out what is going on. Something
related to a pending read I suspect.

Mark

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

Re: Connector refactoring progress update

[Rémy Maucherat <re...@apache.org>]

2015-01-15 11:11 GMT+01:00 Mark Thomas <ma...@apache.org>:

> The latest series of commits is the prep work for merging the
> Http11InputBuffer implementations. I'll be doing that shortly.
>
> The current status is that the tests pass but performance is currently
> worse for all connectors but especially for NIO2.
>

I just checked and my ab doesn't complete for whatever reason on NIO2 (the
very last request hangs), but with the right timing to kiil it, I don't
think the throughput is very much changed.

>
> Once I complete the InputBuffer merging, I plan on spending some time
> looking at tuning. I suspect that the buffering is not as efficient as
> it could be.
>
> Ok sure.

Rémy