[Axis2] Performance

[Glen Daniels <gl...@thoughtcraft.com>]

Hi folks:

Do we have current performance numbers for Axis2?   Ideally a small 
table with timing and memory footprint for 10, 100, 1000, 100000 
identical WS round trips for Axis2, Axis1, and perhaps Glassfish would 
be perfect.  Anyone doing this sort of thing yet?

Do we indeed know that Axis2 is really that much faster/smaller than 
Axis1 at this point?

Thanks,
--Glen

Re: [Axis2] Performance

[Sanjiva Weerawarana <sa...@opensource.lk>]

On Tue, 2006-03-28 at 08:52 -0500, Glen Daniels wrote:
> Hi folks:
> 
> Do we have current performance numbers for Axis2?   Ideally a small 
> table with timing and memory footprint for 10, 100, 1000, 100000 
> identical WS round trips for Axis2, Axis1, and perhaps Glassfish would 
> be perfect.  Anyone doing this sort of thing yet?

Thanks for volunteering Glen :-).

> Do we indeed know that Axis2 is really that much faster/smaller than 
> Axis1 at this point?

Yes, but we need more benchmarks ..

Sanjiva.

Re: [Axis2] Performance

[Glen Daniels <gl...@thoughtcraft.com>]

Hi Dennis!

> I've got a test with Axiom which I'll be adding to Jira later today. If 
> you don't expand the tree Axiom does well (naturally), but if you do 
> it's slower and bulkier than any standard document models. The output is 
> surprisingly slow, too - I'm not sure if this is because of the use of 
> XMLStreamWriter or because of Axiom's way of interfacing to 
> XMLStreamWriter.

Looking forward to that.  I don't think that it's acceptable that Axiom 
in fully expanded form is bigger/bulkier than DOM or JDOM in fully 
expanded form.  We should try to work on that.  Ditto for slow output 
speeds.  Once we have a solid test base, we can experiment with 
tightening up the implementation in various ways....

--Glen

Re: [Axis2] Performance

[Dennis Sosnoski <dm...@sosnoski.com>]

Hi Glen,

I'll be updating my earthquake data retrieval tests over the next week 
or two to bring them up to date, including JiBX binding as an 
alternative. Last time I checked the test running XMLBeans was roughly 
on a par with Axis1 when I modified the code to only use the XMLBeans 
objects for input and output (not as part of the loops). ADB was not 
working well enough to test last time I tried (December). The JiBX 
binding should be considerably faster, especially since I kludge the OM 
interface to avoid expanding the tree when MTOM tries to walk it on output.

I've got a test with Axiom which I'll be adding to Jira later today. If 
you don't expand the tree Axiom does well (naturally), but if you do 
it's slower and bulkier than any standard document models. The output is 
surprisingly slow, too - I'm not sure if this is because of the use of 
XMLStreamWriter or because of Axiom's way of interfacing to XMLStreamWriter.

  - Dennis

Glen Daniels wrote:

> Hi folks:
>
> Do we have current performance numbers for Axis2?   Ideally a small 
> table with timing and memory footprint for 10, 100, 1000, 100000 
> identical WS round trips for Axis2, Axis1, and perhaps Glassfish would 
> be perfect.  Anyone doing this sort of thing yet?
>
> Do we indeed know that Axis2 is really that much faster/smaller than 
> Axis1 at this point?
>
> Thanks,
> --Glen
>