EJB HTTP protocol

[Q Beukes]

Hey,

I read a thread on OpenEJBs performance (http://is.gd/4bMYZ), and
that's a major improvement, ending up in truly great performance. I
don't believe in comparative benchmarks as a "say-it-all judge-all",
but when I had performance issues on Glassfish I was looking for ways
to strip it down to make it faster, and in the process found that
people weren't even getting above 1000 txs/second.And these were with
empty session beans, purely testing the creation/destruction and
turnaround of the remote EJB protocol/setup.

I just want to confirm something. If I find I need the performance
increase when my clients increase to close to/over a hundred, I should
change the java.naming.provider.url property to
"http://localhost:4204/ejb", from the alternative
"ejbd://localhost:4201" ? This will then be what is mentioned in the
thread, ie. the http over jetty code setup?

Quintin Beukes

[David Blevins]

On Oct 10, 2009, at 3:27 AM, Quintin Beukes wrote:

> Hey,
>
> I read a thread on OpenEJBs performance (http://is.gd/4bMYZ), and
> that's a major improvement, ending up in truly great performance. I
> don't believe in comparative benchmarks as a "say-it-all judge-all",
> but when I had performance issues on Glassfish I was looking for ways
> to strip it down to make it faster, and in the process found that
> people weren't even getting above 1000 txs/second.And these were with
> empty session beans, purely testing the creation/destruction and
> turnaround of the remote EJB protocol/setup.
>
> I just want to confirm something. If I find I need the performance
> increase when my clients increase to close to/over a hundred, I should
> change the java.naming.provider.url property to
> "http://localhost:4204/ejb", from the alternative
> "ejbd://localhost:4201" ? This will then be what is mentioned in the
> thread, ie. the http over jetty code setup?

Right.  That and you need to add the jetty jars to lib/.  We will  
detect it on startup and magically use the jetty http impl instead of  
our own http impl which is really just for testing.

-David

[Quintin Beukes-2]

Is this documented on the web site - just want to determine if I can
delete this mail or archive it?

Quintin Beukes



On Thu, Oct 15, 2009 at 3:58 AM, David Blevins <david.blevins@...> wrote:

>
> On Oct 10, 2009, at 3:27 AM, Quintin Beukes wrote:
>
>> Hey,
>>
>> I read a thread on OpenEJBs performance (http://is.gd/4bMYZ), and
>> that's a major improvement, ending up in truly great performance. I
>> don't believe in comparative benchmarks as a "say-it-all judge-all",
>> but when I had performance issues on Glassfish I was looking for ways
>> to strip it down to make it faster, and in the process found that
>> people weren't even getting above 1000 txs/second.And these were with
>> empty session beans, purely testing the creation/destruction and
>> turnaround of the remote EJB protocol/setup.
>>
>> I just want to confirm something. If I find I need the performance
>> increase when my clients increase to close to/over a hundred, I should
>> change the java.naming.provider.url property to
>> "http://localhost:4204/ejb", from the alternative
>> "ejbd://localhost:4201" ? This will then be what is mentioned in the
>> thread, ie. the http over jetty code setup?
>
> Right.  That and you need to add the jetty jars to lib/.  We will detect it
> on startup and magically use the jetty http impl instead of our own http
> impl which is really just for testing.
>
> -David
>
>

[David Blevins]

Archive it.  Or if you get a cycle to help add to the docs (or blog  
about it), that helps too :)

-David

On Oct 19, 2009, at 11:53 AM, Quintin Beukes wrote:

> Is this documented on the web site - just want to determine if I can
> delete this mail or archive it?
>
> Quintin Beukes
>
>
>
> On Thu, Oct 15, 2009 at 3:58 AM, David Blevins  
> <david.blevins@...> wrote:
>>
>> On Oct 10, 2009, at 3:27 AM, Quintin Beukes wrote:
>>
>>> Hey,
>>>
>>> I read a thread on OpenEJBs performance (http://is.gd/4bMYZ), and
>>> that's a major improvement, ending up in truly great performance. I
>>> don't believe in comparative benchmarks as a "say-it-all judge-all",
>>> but when I had performance issues on Glassfish I was looking for  
>>> ways
>>> to strip it down to make it faster, and in the process found that
>>> people weren't even getting above 1000 txs/second.And these were  
>>> with
>>> empty session beans, purely testing the creation/destruction and
>>> turnaround of the remote EJB protocol/setup.
>>>
>>> I just want to confirm something. If I find I need the performance
>>> increase when my clients increase to close to/over a hundred, I  
>>> should
>>> change the java.naming.provider.url property to
>>> "http://localhost:4204/ejb", from the alternative
>>> "ejbd://localhost:4201" ? This will then be what is mentioned in the
>>> thread, ie. the http over jetty code setup?
>>
>> Right.  That and you need to add the jetty jars to lib/.  We will  
>> detect it
>> on startup and magically use the jetty http impl instead of our own  
>> http
>> impl which is really just for testing.
>>
>> -David
>>
>>
>